docs: 补充回测结果摘要 + 策略文档 + Dashboard 后端 + 整理目录结构
This commit is contained in:
14
README.md
14
README.md
@ -24,6 +24,20 @@ ETH/USDT 永续合约量化交易策略版本管理，基于 freqtrade + Binance
|
||||
- v2.2b：当前最优回测基线（+4673%/+17%最大回撤）
|
||||
- v2.2d：D1趋势总闸门 — 震荡市不下单是保护机制不是bug
|
||||
|
||||
## 仓库结构
|
||||
|
||||
```
|
||||
.
|
||||
├── strategy.py # 当前最新策略（v2.2d）
|
||||
├── indicators.py # Dashboard 后端指标计算
|
||||
├── config.backtest.json # 标准化回测配置
|
||||
├── ablation/ # v2.1 消融实验（8组 + 基线）
|
||||
├── backtest/ # 关键回测结果摘要
|
||||
├── docs/ # 策略白皮书、技术文档
|
||||
├── dashboard/ # Dashboard 后端
|
||||
└── configs/ # 各版本配置文件
|
||||
```
|
||||
|
||||
## 铁律
|
||||
|
||||
1. 只增不删 — 所有历史版本保留
|
||||
|
||||
39
backtest/INDEX.md
Normal file
39
backtest/INDEX.md
Normal file
@ -0,0 +1,39 @@
|
||||
# Backtest Results Index
|
||||
|
||||
## Today's Backtests (2026-06-08)
|
||||
|
||||
| 文件 | 策略 | 时间范围 | 周期 | 大小 | 备注 |
|
||||
|------|------|----------|------|------|------|
|
||||
| backtest-result-2026-06-08_00-04-25.meta.json | StructureFlowStrategyV16 | 2022-01-01~2025-08-17 | 2022-01-01~2025-08-17 | 511KB | ✅ 基线版本 |
|
||||
| backtest-result-2026-06-08_00-04-50.meta.json | StructureFlowStrategyV16 | 2022-01-01~2025-08-17 | 2022-01-01~2025-08-17 | 517KB | ✅ 基线版本 |
|
||||
| backtest-result-2026-06-08_00-08-01.meta.json | StructureFlowStrategyV16 | 2022-01-01~2025-08-17 | 2022-01-01~2025-08-17 | 511KB | ✅ 基线版本 |
|
||||
| backtest-result-2026-06-08_00-10-04.meta.json | StructureFlowStrategyV16 | 2022-01-01~2025-08-17 | 2022-01-01~2025-08-17 | 517KB | ✅ 基线版本 |
|
||||
| backtest-result-2026-06-08_00-10-14.meta.json | StructureFlowStrategyV16 | 2022-01-01~2025-08-17 | 2022-01-01~2025-08-17 | 511KB | ✅ 基线版本 |
|
||||
| backtest-result-2026-06-08_00-07-46.meta.json | StructureFlowStrategyV16 | 2022-01-01~2023-01-01 | 2022年度 | 149KB | ✅ 基线版本 |
|
||||
| backtest-result-2026-06-08_00-07-45.meta.json | StructureFlowStrategyV16 | 2023-01-01~2024-01-01 | 2023年度 | 146KB | ✅ 基线版本 |
|
||||
| backtest-result-2026-06-08_00-07-48.meta.json | StructureFlowStrategyV16 | 2023-01-01~2024-01-01 | 2023年度 | 143KB | ✅ 基线版本 |
|
||||
| backtest-result-2026-06-08_00-07-49.meta.json | StructureFlowStrategyV16 | 2024-01-01~2025-01-01 | 2024年度 | 149KB | ✅ 基线版本 |
|
||||
| backtest-result-2026-06-08_00-00-50.meta.json | StructureFlowStrategyV16 | 2025-01-01~2025-08-17 | 2025-01-01~2025-08-17 | 98KB | ✅ 基线版本 |
|
||||
| backtest-result-2026-06-08_00-09-34.meta.json | StructureFlowStrategyV16 | 2025-01-01~2025-08-17 | 2025-01-01~2025-08-17 | 99KB | ✅ 基线版本 |
|
||||
| backtest-result-2026-06-08_00-09-36.meta.json | StructureFlowStrategyV16 | 2025-01-01~2025-08-17 | 2025-01-01~2025-08-17 | 98KB | ✅ 基线版本 |
|
||||
| backtest-result-2026-06-08_00-25-02.meta.json | StructureFlowStrategyV18 | 2022-01-01~2026-06-07 | 全周期 | 611KB | ⚠️ 对比用 (2%缓冲) |
|
||||
| backtest-result-2026-06-08_00-25-05.meta.json | StructureFlowStrategyV18 | 2022-01-01~2026-06-07 | 全周期 | 615KB | ⚠️ 对比用 (2%缓冲) |
|
||||
| backtest-result-2026-06-08_00-17-35.meta.json | StructureFlowStrategyV17 | 2022-01-01~2026-06-07 | 全周期 | 607KB | ❌ 失败 (止损太宽) |
|
||||
| backtest-result-2026-06-08_14-50-07.meta.json | StructureFlowStrategyV161 | 2022-01-01~2026-06-07 | 全周期 | - | ❌ v1.6.1 过度过滤 |
|
||||
| backtest-result-2026-06-08_14-50-41.meta.json | StructureFlowStrategyV161 | 2022-01-01~2026-06-07 | 全周期 | - | ❌ v1.6.1 过度过滤 |
|
||||
| backtest-result-2026-06-08_15-07-41.meta.json | StructureFlowStrategyV162 | 2022-01-01~2026-06-07 | 全周期 | - | ❌ v1.6.2 Brooks二次确认 |
|
||||
| backtest-result-2026-06-08_07-22-32.zip | StructureFlowStrategyV163 | 2022-01-01~2026-06-07 | 全周期 | 622KB | ❌ v1.6.3 H4过滤器误杀盈利 |
|
||||
| backtest-result-2026-06-08_08-45-17.zip | StructureFlowStrategyV21 | 2022-01-01~2026-06-07 | 全周期 (ETH, Binance Futures) | 627KB | ⭐ v2.1 D1趋势强度 ETH+304.31% |
|
||||
| backtest-result-2026-06-08_08-46-37.zip | StructureFlowStrategyV21 | 2022-01-01~2026-06-07 | 全周期 (BTC, Binance Futures) | 632KB | ⭐ v2.1 D1趋势强度 BTC+95.96% |
|
||||
| backtest-result-2026-06-08_08-45-58.zip | StructureFlowStrategyV16 | 2022-01-01~2026-06-07 | 全周期 (ETH, Binance Futures) | 628KB | v1.6 基线 ETH+282.27% (同环境对比) |
|
||||
| backtest-result-2026-06-08_08-47-14.zip | StructureFlowStrategyV16 | 2022-01-01~2026-06-07 | 全周期 (BTC, Binance Futures) | 629KB | v1.6 基线 BTC+90.92% (同环境对比) |
|
||||
|
||||
**v2.1 vs v1.6 同环境对比 (Binance Futures, 2022-2026):**
|
||||
- ETH: v1.6 +282.27% → v2.1 +304.31% (+7.8%相对提升)
|
||||
- BTC: v1.6 +90.92% → v2.1 +95.96% (+5.5%相对提升)
|
||||
- 两者PF/CAGR/DD均改善，v2.1全面超越v1.6
|
||||
|
||||
**⚠️ 数据源说明：**
|
||||
- 之前的v1.6回测(3659.63%)用的是OKX Spot数据，fee结构不同
|
||||
- 今日v2.1回测用的是Binance Futures数据（0.05% fee）
|
||||
- 两种数据源下v2.1都超越v1.6，趋势一致
|
||||
File diff suppressed because one or more lines are too long
@ -0,0 +1,435 @@
|
||||
"""
|
||||
Structure Flow Strategy v1.6
|
||||
=======================
|
||||
变更记录：
|
||||
v1.0 (2026-06-07): 纯价格结构策略，D1定方向→4H定位→1H入场
|
||||
v1.1 (2026-06-07): 1H futures，结构止损，首次回测成功(+61.52%)
|
||||
v1.2 (2026-06-07): Entry Candle止损，bug导致50笔硬止损全亏
|
||||
v1.3 (2026-06-07): ATR动态止损，结果-63.72%，胜率20.2%
|
||||
v1.4 (2026-06-07): 回归纯价格结构止损，+140.71%，胜率38.7%
|
||||
v1.5 (2026-06-07): 参数调优(stoploss -5%→-15%, max_stop_dist 3%→5%)，+140.83%
|
||||
v1.6 (2026-06-07): ===== 入场质量优化 =====
|
||||
- 6-bar冷却期：信号后6h内不重复入场（防止连挨多刀）
|
||||
- 活支撑/阻力检查：S/R必须被最近测试并守住才算有效
|
||||
设计原则：不降频，只砍最差的那几笔重复入场
|
||||
"""
|
||||
|
||||
from datetime import datetime
|
||||
import numpy as np
|
||||
import pandas as pd
|
||||
from pandas import DataFrame
|
||||
from freqtrade.strategy import IStrategy, IntParameter, informative
|
||||
from freqtrade.persistence import Trade
|
||||
|
||||
|
||||
class StructureFlowStrategyV16(IStrategy):
|
||||
"""
|
||||
Structure Flow Strategy v1.6 — 纯价格结构，零指标
|
||||
|
||||
v1.6改动（相对于v1.5）：
|
||||
1. 6-bar冷却期：同方向信号触发后，6h内禁止同向再入场
|
||||
→ 解决"同一天同一个价位挨两刀"的问题
|
||||
2. 活支撑/阻力检查：4H Swing Point 必须被价格测试并守住才有效
|
||||
→ 解决"在死支撑上入场"的问题
|
||||
"""
|
||||
|
||||
can_short = True
|
||||
stoploss = -0.15
|
||||
use_custom_stoploss = True
|
||||
minimal_roi = {"0": 100}
|
||||
max_open_trades = 1
|
||||
timeframe = "1h"
|
||||
|
||||
# =====================
|
||||
# 可优化参数
|
||||
# =====================
|
||||
|
||||
swing_lookback_d1 = IntParameter(8, 14, default=10, space="buy")
|
||||
swing_lookback_h4 = IntParameter(5, 10, default=8, space="buy")
|
||||
pin_bar_wick_ratio = IntParameter(50, 70, default=60, space="buy")
|
||||
max_stop_dist = IntParameter(20, 50, default=50, space="buy")
|
||||
# v1.6 新增
|
||||
cooldown_bars = IntParameter(3, 12, default=6, space="buy")
|
||||
|
||||
# =====================
|
||||
# 工具：Swing Point 检测
|
||||
# =====================
|
||||
|
||||
@staticmethod
|
||||
def _detect_swing_points(
|
||||
high: pd.Series,
|
||||
low: pd.Series,
|
||||
window: int = 5,
|
||||
) -> tuple[pd.Series, pd.Series]:
|
||||
n = len(high)
|
||||
sh = pd.Series(np.nan, index=high.index, dtype=float)
|
||||
sl = pd.Series(np.nan, index=low.index, dtype=float)
|
||||
|
||||
for i in range(window, n - window):
|
||||
if high.iloc[i] > high.iloc[i - window : i].max() and high.iloc[i] > high.iloc[i + 1 : i + window + 1].max():
|
||||
sh.iloc[i] = high.iloc[i]
|
||||
if low.iloc[i] < low.iloc[i - window : i].min() and low.iloc[i] < low.iloc[i + 1 : i + window + 1].min():
|
||||
sl.iloc[i] = low.iloc[i]
|
||||
|
||||
return sh, sl
|
||||
|
||||
# =====================
|
||||
# 工具：结构分析
|
||||
# =====================
|
||||
|
||||
def _build_structure(
|
||||
self,
|
||||
high: pd.Series,
|
||||
low: pd.Series,
|
||||
close: pd.Series,
|
||||
swing_high: pd.Series,
|
||||
swing_low: pd.Series,
|
||||
) -> DataFrame:
|
||||
n = len(high)
|
||||
|
||||
trend_up_arr = np.full(n, False)
|
||||
trend_down_arr = np.full(n, False)
|
||||
nearest_support = np.full(n, np.nan)
|
||||
nearest_resistance = np.full(n, np.nan)
|
||||
in_demand_zone = np.full(n, False)
|
||||
in_supply_zone = np.full(n, False)
|
||||
|
||||
sh_prices = []
|
||||
sl_prices = []
|
||||
|
||||
for i in range(n):
|
||||
if pd.notna(swing_high.iloc[i]):
|
||||
sh_prices.append(swing_high.iloc[i])
|
||||
if len(sh_prices) > 4:
|
||||
sh_prices.pop(0)
|
||||
|
||||
if pd.notna(swing_low.iloc[i]):
|
||||
sl_prices.append(swing_low.iloc[i])
|
||||
if len(sl_prices) > 4:
|
||||
sl_prices.pop(0)
|
||||
|
||||
if len(sh_prices) >= 2 and len(sl_prices) >= 2:
|
||||
if sh_prices[-1] > sh_prices[-2] and sl_prices[-1] > sl_prices[-2]:
|
||||
trend_up_arr[i] = True
|
||||
elif sh_prices[-1] < sh_prices[-2] and sl_prices[-1] < sl_prices[-2]:
|
||||
trend_down_arr[i] = True
|
||||
elif i > 0:
|
||||
trend_up_arr[i] = trend_up_arr[i - 1]
|
||||
trend_down_arr[i] = trend_down_arr[i - 1]
|
||||
elif i > 0:
|
||||
trend_up_arr[i] = trend_up_arr[i - 1]
|
||||
trend_down_arr[i] = trend_down_arr[i - 1]
|
||||
|
||||
if sl_prices:
|
||||
nearest_support[i] = sl_prices[-1]
|
||||
if sh_prices:
|
||||
nearest_resistance[i] = sh_prices[-1]
|
||||
|
||||
c = close.iloc[i]
|
||||
if not np.isnan(nearest_support[i]) and not np.isnan(nearest_resistance[i]):
|
||||
zone_range = nearest_resistance[i] - nearest_support[i]
|
||||
if zone_range > 0:
|
||||
pos_pct = (c - nearest_support[i]) / zone_range
|
||||
in_demand_zone[i] = pos_pct < 0.35
|
||||
in_supply_zone[i] = pos_pct > 0.65
|
||||
|
||||
return DataFrame({
|
||||
"trend_up": trend_up_arr,
|
||||
"trend_down": trend_down_arr,
|
||||
"support": nearest_support,
|
||||
"resistance": nearest_resistance,
|
||||
"in_demand": in_demand_zone,
|
||||
"in_supply": in_supply_zone,
|
||||
}, index=high.index)
|
||||
|
||||
# =====================
|
||||
# 工具：K线形态检测
|
||||
# =====================
|
||||
|
||||
@staticmethod
|
||||
def _detect_candle_patterns(
|
||||
open_: pd.Series,
|
||||
high: pd.Series,
|
||||
low: pd.Series,
|
||||
close: pd.Series,
|
||||
pin_bar_wick_ratio: float = 0.6,
|
||||
) -> tuple[pd.Series, pd.Series, pd.Series, pd.Series]:
|
||||
body = (close - open_).abs()
|
||||
total_range = (high - low).replace(0, 0.0001)
|
||||
|
||||
upper_wick = high - close.where(close > open_, open_)
|
||||
lower_wick = open_.where(close > open_, close) - low
|
||||
is_pin = (upper_wick + lower_wick) / total_range > pin_bar_wick_ratio
|
||||
|
||||
bullish_pin = is_pin & (close > open_) & (lower_wick > upper_wick)
|
||||
bearish_pin = is_pin & (close < open_) & (upper_wick > lower_wick)
|
||||
|
||||
prev_open = open_.shift(1)
|
||||
prev_close = close.shift(1)
|
||||
bullish_engulf = (close > prev_open) & (open_ < prev_close) & (close > open_)
|
||||
bearish_engulf = (close < prev_open) & (open_ > prev_close) & (close < open_)
|
||||
|
||||
return bullish_pin, bearish_pin, bullish_engulf, bearish_engulf
|
||||
|
||||
# ================================================================
|
||||
# 信息时间框架 — D1 宏观结构
|
||||
# ================================================================
|
||||
|
||||
@informative("1d")
|
||||
def populate_indicators_1d(
|
||||
self, dataframe: DataFrame, metadata: dict
|
||||
) -> DataFrame:
|
||||
sh, sl = self._detect_swing_points(
|
||||
dataframe["high"], dataframe["low"],
|
||||
self.swing_lookback_d1.value,
|
||||
)
|
||||
structure = self._build_structure(
|
||||
dataframe["high"], dataframe["low"], dataframe["close"],
|
||||
sh, sl,
|
||||
)
|
||||
dataframe["trend_up"] = structure["trend_up"]
|
||||
dataframe["trend_down"] = structure["trend_down"]
|
||||
return dataframe
|
||||
|
||||
# ================================================================
|
||||
# 信息时间框架 — 4H 中期结构
|
||||
# ================================================================
|
||||
|
||||
@informative("4h")
|
||||
def populate_indicators_4h(
|
||||
self, dataframe: DataFrame, metadata: dict
|
||||
) -> DataFrame:
|
||||
sh, sl = self._detect_swing_points(
|
||||
dataframe["high"], dataframe["low"],
|
||||
self.swing_lookback_h4.value,
|
||||
)
|
||||
structure = self._build_structure(
|
||||
dataframe["high"], dataframe["low"], dataframe["close"],
|
||||
sh, sl,
|
||||
)
|
||||
dataframe["trend_up"] = structure["trend_up"]
|
||||
dataframe["trend_down"] = structure["trend_down"]
|
||||
dataframe["support"] = structure["support"]
|
||||
dataframe["resistance"] = structure["resistance"]
|
||||
dataframe["in_demand"] = structure["in_demand"]
|
||||
dataframe["in_supply"] = structure["in_supply"]
|
||||
|
||||
# ================================
|
||||
# v1.6 新增：活支撑/阻力检查
|
||||
# ================================
|
||||
# 支撑"活"的条件：在最近3根4H bar内，low 触及 support ±0.5%
|
||||
# 并且收盘价在支撑之上（即测试后撑住了）
|
||||
touched_support = (
|
||||
(dataframe["low"] <= dataframe["support"] * 1.005) &
|
||||
(dataframe["low"] >= dataframe["support"] * 0.995)
|
||||
)
|
||||
held_support = dataframe["close"] > dataframe["support"]
|
||||
support_tested_and_held = touched_support & held_support
|
||||
# 在过去3根4H bar内有至少一次"测试并守住"
|
||||
dataframe["support_alive"] = support_tested_and_held.rolling(3, min_periods=1).max() > 0
|
||||
|
||||
# 阻力"活"的条件：high 触及 resistance ±0.5% 且 close 在阻力之下
|
||||
touched_resistance = (
|
||||
(dataframe["high"] >= dataframe["resistance"] * 0.995) &
|
||||
(dataframe["high"] <= dataframe["resistance"] * 1.005)
|
||||
)
|
||||
held_resistance = dataframe["close"] < dataframe["resistance"]
|
||||
resistance_tested_and_held = touched_resistance & held_resistance
|
||||
dataframe["resistance_alive"] = resistance_tested_and_held.rolling(3, min_periods=1).max() > 0
|
||||
|
||||
return dataframe
|
||||
|
||||
# ================================================================
|
||||
# 主时间框架 — 1H 指标
|
||||
# ================================================================
|
||||
|
||||
def populate_indicators(
|
||||
self, dataframe: DataFrame, metadata: dict
|
||||
) -> DataFrame:
|
||||
"""1H 级别：K线形态（零指标）。"""
|
||||
bullish_pin, bearish_pin, bullish_engulf, bearish_engulf = (
|
||||
self._detect_candle_patterns(
|
||||
dataframe["open"],
|
||||
dataframe["high"],
|
||||
dataframe["low"],
|
||||
dataframe["close"],
|
||||
self.pin_bar_wick_ratio.value / 100.0,
|
||||
)
|
||||
)
|
||||
dataframe["bullish_pinbar"] = bullish_pin
|
||||
dataframe["bearish_pinbar"] = bearish_pin
|
||||
dataframe["bullish_engulfing"] = bullish_engulf
|
||||
dataframe["bearish_engulfing"] = bearish_engulf
|
||||
dataframe["bullish_signal"] = bullish_pin | bullish_engulf
|
||||
dataframe["bearish_signal"] = bearish_pin | bearish_engulf
|
||||
|
||||
# NaN 安全处理
|
||||
bool_cols = [
|
||||
"trend_up_1d", "trend_down_1d",
|
||||
"trend_up_4h", "trend_down_4h",
|
||||
"in_demand_4h", "in_supply_4h",
|
||||
"support_alive_4h", "resistance_alive_4h",
|
||||
"bullish_signal", "bearish_signal",
|
||||
]
|
||||
for col in bool_cols:
|
||||
if col in dataframe.columns:
|
||||
dataframe[col] = dataframe[col].fillna(False)
|
||||
|
||||
return dataframe
|
||||
|
||||
# =====================
|
||||
# 入场信号
|
||||
# =====================
|
||||
|
||||
def populate_entry_trend(self, dataframe: DataFrame, metadata: dict) -> DataFrame:
|
||||
"""
|
||||
入场逻辑（1H 时间框架）。
|
||||
|
||||
做多条件:
|
||||
1. D1 上升结构（trend_up_1d）
|
||||
2. 4H 需求区域（in_demand_4h）
|
||||
3. 1H 看涨 K 线形态（bullish_signal）
|
||||
4. 止损距离 ≤ max_stop_dist%
|
||||
5. [v1.6] 支撑位是"活"的（support_alive_4h）
|
||||
6. [v1.6] 6h内没有过同方向入场信号（冷却期）
|
||||
|
||||
做空条件对称。
|
||||
"""
|
||||
max_dist = self.max_stop_dist.value / 100.0
|
||||
cooldown = self.cooldown_bars.value
|
||||
|
||||
# NaN 安全处理
|
||||
bool_cols = [
|
||||
"trend_up_1d", "trend_down_1d",
|
||||
"trend_up_4h", "trend_down_4h",
|
||||
"in_demand_4h", "in_supply_4h",
|
||||
"support_alive_4h", "resistance_alive_4h",
|
||||
"bullish_signal", "bearish_signal",
|
||||
]
|
||||
for col in bool_cols:
|
||||
if col in dataframe.columns:
|
||||
dataframe[col] = dataframe[col].fillna(False)
|
||||
|
||||
# ── 做多 ──
|
||||
long_stop_dist = (dataframe["open"] - dataframe["support_4h"]) / dataframe["open"]
|
||||
|
||||
long_base = (
|
||||
dataframe["trend_up_1d"]
|
||||
& dataframe["in_demand_4h"]
|
||||
& dataframe["bullish_signal"]
|
||||
& (long_stop_dist <= max_dist)
|
||||
& (long_stop_dist > 0.003)
|
||||
)
|
||||
|
||||
# v1.6: 活支撑 — 支撑必须在最近3根4H内被测试并守住
|
||||
long_base = long_base & dataframe["support_alive_4h"]
|
||||
|
||||
# v1.6: 冷却期 — 过去N根1H bar内没有过满足条件的做多信号
|
||||
long_recent = long_base.rolling(cooldown, min_periods=1).max().shift(1) == 0
|
||||
long_conditions = long_base & long_recent
|
||||
|
||||
dataframe.loc[long_conditions, "enter_long"] = 1
|
||||
|
||||
# ── 做空 ──
|
||||
short_stop_dist = (dataframe["resistance_4h"] - dataframe["open"]) / dataframe["open"]
|
||||
|
||||
short_base = (
|
||||
dataframe["trend_down_1d"]
|
||||
& dataframe["in_supply_4h"]
|
||||
& dataframe["bearish_signal"]
|
||||
& (short_stop_dist <= max_dist)
|
||||
& (short_stop_dist > 0.003)
|
||||
)
|
||||
|
||||
# v1.6: 活阻力 — 阻力必须在最近3根4H内被测试并守住
|
||||
short_base = short_base & dataframe["resistance_alive_4h"]
|
||||
|
||||
# v1.6: 冷却期 — 过去N根1H bar内没有过满足条件的做空信号
|
||||
short_recent = short_base.rolling(cooldown, min_periods=1).max().shift(1) == 0
|
||||
short_conditions = short_base & short_recent
|
||||
|
||||
dataframe.loc[short_conditions, "enter_short"] = 1
|
||||
|
||||
return dataframe
|
||||
|
||||
# =====================
|
||||
# 出场信号
|
||||
# =====================
|
||||
|
||||
def populate_exit_trend(self, dataframe: DataFrame, metadata: dict) -> DataFrame:
|
||||
"""出场逻辑 — 由结构反转触发。"""
|
||||
exit_long = ~dataframe["trend_up_1d"].fillna(True)
|
||||
dataframe.loc[exit_long, "exit_long"] = 1
|
||||
|
||||
exit_short = dataframe["trend_up_1d"].fillna(False)
|
||||
dataframe.loc[exit_short, "exit_short"] = 1
|
||||
|
||||
return dataframe
|
||||
|
||||
# =====================
|
||||
# 动态止损 — 纯价格结构（基于Swing Point）
|
||||
# =====================
|
||||
|
||||
def custom_stoploss(
|
||||
self,
|
||||
pair: str,
|
||||
trade: Trade,
|
||||
current_time: datetime,
|
||||
current_rate: float,
|
||||
current_profit: float,
|
||||
after_fill: bool,
|
||||
**kwargs,
|
||||
) -> float:
|
||||
"""
|
||||
止损逻辑：完全基于价格结构，零指标。
|
||||
|
||||
止损位：
|
||||
做多 → support_4h - 0.1%缓冲（最近4H Swing Low下方）
|
||||
做空 → resistance_4h + 0.1%缓冲（最近4H Swing High上方）
|
||||
|
||||
support_4h / resistance_4h 随新Swing Point自动更新，
|
||||
天然形成追踪止损效果。
|
||||
"""
|
||||
dataframe, _ = self.dp.get_analyzed_dataframe(pair, self.timeframe)
|
||||
if dataframe is None or len(dataframe) == 0:
|
||||
return -0.02 if not trade.is_short else 0.02
|
||||
|
||||
last = dataframe.iloc[-1]
|
||||
|
||||
if not trade.is_short:
|
||||
support = last.get("support_4h", np.nan)
|
||||
if pd.isna(support) or support <= 0:
|
||||
return -0.02
|
||||
sl_price = support * 0.999
|
||||
sl_ratio = (sl_price / current_rate) - 1.0
|
||||
return max(sl_ratio, -0.15)
|
||||
else:
|
||||
resistance = last.get("resistance_4h", np.nan)
|
||||
if pd.isna(resistance) or resistance <= 0:
|
||||
return 0.02
|
||||
sl_price = resistance * 1.001
|
||||
sl_ratio = 1.0 - (sl_price / current_rate)
|
||||
return min(sl_ratio, 0.15)
|
||||
|
||||
# =====================
|
||||
# Plot config
|
||||
# =====================
|
||||
|
||||
@staticmethod
|
||||
def plot_config() -> dict:
|
||||
return {
|
||||
"main_plot": {
|
||||
"support_4h": {"color": "green", "type": "line"},
|
||||
"resistance_4h": {"color": "red", "type": "line"},
|
||||
},
|
||||
"subplots": {
|
||||
"signals": {
|
||||
"bullish_pinbar": {"color": "green", "type": "scatter"},
|
||||
"bearish_pinbar": {"color": "red", "type": "scatter"},
|
||||
},
|
||||
"filters": {
|
||||
"support_alive_4h": {"color": "green", "type": "line"},
|
||||
"resistance_alive_4h": {"color": "red", "type": "line"},
|
||||
},
|
||||
},
|
||||
}
|
||||
@ -0,0 +1 @@
|
||||
{"max_open_trades":1,"stake_currency":"USDT","stake_amount":"unlimited","tradable_balance_ratio":0.99,"fiat_display_currency":"USD","dry_run":true,"dry_run_wallet":10000,"trading_mode":"futures","margin_mode":"cross","liquidation_buffer":0.05,"exchange":{"name":"binance","key":"REDACTED","secret":"REDACTED","password":"REDACTED","ccxt_config":{"enableRateLimit":true},"ccxt_async_config":{"enableRateLimit":true},"pair_whitelist":["ETH/USDT:USDT"],"pair_blacklist":[]},"pairlists":[{"method":"StaticPairList"}],"telegram":{"enabled":false,"token":"REDACTED","chat_id":"REDACTED"},"api_server":{"enabled":false,"listen_ip_address":"0.0.0.0","listen_port":8080,"username":"freqtrader","password":"REDACTED","jwt_secret_key":"somethingRandom123"},"bot_name":"backtest","entry_pricing":{"price_side":"same","use_order_book":true,"order_book_top":1,"price_last_balance":0.0,"check_depth_of_market":{"enabled":false,"bids_to_ask_delta":1}},"exit_pricing":{"price_side":"same","use_order_book":true,"order_book_top":1},"config_files":["user_data/config_backtest.json"],"internals":{}}
|
||||
Binary file not shown.
File diff suppressed because one or more lines are too long
@ -0,0 +1,456 @@
|
||||
"""
|
||||
Structure Flow Strategy v2.1
|
||||
=======================
|
||||
变更记录：
|
||||
v1.6 (2026-06-07): 最优基线 — +3659.63%, 190笔, 69.3% trailing胜率
|
||||
v2.0 (2026-06-08): B1 入场延迟确认 — 方向正确但降频严重
|
||||
v2.1 (2026-06-08): ===== D1: 趋势强度过滤 =====
|
||||
在4H级别评估趋势强度：最近2个Swing Point的间距变化。
|
||||
如果趋势在扩张（HH/HL间距增大），允许入场；
|
||||
如果趋势在收缩（HH/HL间距缩小或震荡），过滤信号。
|
||||
目的：只在趋势明确时交易，避免震荡市反复止损。
|
||||
"""
|
||||
|
||||
from datetime import datetime
|
||||
import numpy as np
|
||||
import pandas as pd
|
||||
from pandas import DataFrame
|
||||
from freqtrade.strategy import IStrategy, IntParameter, informative
|
||||
from freqtrade.persistence import Trade
|
||||
|
||||
|
||||
class StructureFlowStrategyV21(IStrategy):
|
||||
"""
|
||||
Structure Flow Strategy v2.1 — D1: 趋势强度过滤
|
||||
|
||||
v2.1改动（相对于v1.6）：
|
||||
在4H级别计算趋势强度：最近2个Swing High间距 + Swing Low间距的变化。
|
||||
只有趋势在扩张（或至少不收缩）时才允许入场。
|
||||
"""
|
||||
|
||||
can_short = True
|
||||
stoploss = -0.15
|
||||
use_custom_stoploss = True
|
||||
minimal_roi = {"0": 100}
|
||||
max_open_trades = 1
|
||||
timeframe = "1h"
|
||||
|
||||
# =====================
|
||||
# 可优化参数
|
||||
# =====================
|
||||
|
||||
swing_lookback_d1 = IntParameter(8, 14, default=10, space="buy")
|
||||
swing_lookback_h4 = IntParameter(5, 10, default=8, space="buy")
|
||||
pin_bar_wick_ratio = IntParameter(50, 70, default=60, space="buy")
|
||||
max_stop_dist = IntParameter(20, 50, default=50, space="buy")
|
||||
cooldown_bars = IntParameter(3, 12, default=6, space="buy")
|
||||
# v2.1 新增：趋势强度最小扩张比例（x/100 = 0%~50%）
|
||||
# 0 = 只要不收缩就行；越大要求趋势扩张越强
|
||||
trend_strength_min = IntParameter(-50, 20, default=-20, space="buy") # 扫描更宽范围
|
||||
|
||||
# =====================
|
||||
# 工具：Swing Point 检测
|
||||
# =====================
|
||||
|
||||
@staticmethod
|
||||
def _detect_swing_points(
|
||||
high: pd.Series,
|
||||
low: pd.Series,
|
||||
window: int = 5,
|
||||
) -> tuple[pd.Series, pd.Series]:
|
||||
n = len(high)
|
||||
sh = pd.Series(np.nan, index=high.index, dtype=float)
|
||||
sl = pd.Series(np.nan, index=low.index, dtype=float)
|
||||
|
||||
for i in range(window, n - window):
|
||||
if high.iloc[i] > high.iloc[i - window : i].max() and high.iloc[i] > high.iloc[i + 1 : i + window + 1].max():
|
||||
sh.iloc[i] = high.iloc[i]
|
||||
if low.iloc[i] < low.iloc[i - window : i].min() and low.iloc[i] < low.iloc[i + 1 : i + window + 1].min():
|
||||
sl.iloc[i] = low.iloc[i]
|
||||
|
||||
return sh, sl
|
||||
|
||||
# =====================
|
||||
# 工具：结构分析
|
||||
# =====================
|
||||
|
||||
def _build_structure(
|
||||
self,
|
||||
high: pd.Series,
|
||||
low: pd.Series,
|
||||
close: pd.Series,
|
||||
swing_high: pd.Series,
|
||||
swing_low: pd.Series,
|
||||
) -> DataFrame:
|
||||
n = len(high)
|
||||
|
||||
trend_up_arr = np.full(n, False)
|
||||
trend_down_arr = np.full(n, False)
|
||||
nearest_support = np.full(n, np.nan)
|
||||
nearest_resistance = np.full(n, np.nan)
|
||||
in_demand_zone = np.full(n, False)
|
||||
in_supply_zone = np.full(n, False)
|
||||
|
||||
sh_prices = []
|
||||
sl_prices = []
|
||||
|
||||
for i in range(n):
|
||||
if pd.notna(swing_high.iloc[i]):
|
||||
sh_prices.append(swing_high.iloc[i])
|
||||
if len(sh_prices) > 4:
|
||||
sh_prices.pop(0)
|
||||
|
||||
if pd.notna(swing_low.iloc[i]):
|
||||
sl_prices.append(swing_low.iloc[i])
|
||||
if len(sl_prices) > 4:
|
||||
sl_prices.pop(0)
|
||||
|
||||
if len(sh_prices) >= 2 and len(sl_prices) >= 2:
|
||||
if sh_prices[-1] > sh_prices[-2] and sl_prices[-1] > sl_prices[-2]:
|
||||
trend_up_arr[i] = True
|
||||
elif sh_prices[-1] < sh_prices[-2] and sl_prices[-1] < sl_prices[-2]:
|
||||
trend_down_arr[i] = True
|
||||
elif i > 0:
|
||||
trend_up_arr[i] = trend_up_arr[i - 1]
|
||||
trend_down_arr[i] = trend_down_arr[i - 1]
|
||||
elif i > 0:
|
||||
trend_up_arr[i] = trend_up_arr[i - 1]
|
||||
trend_down_arr[i] = trend_down_arr[i - 1]
|
||||
|
||||
if sl_prices:
|
||||
nearest_support[i] = sl_prices[-1]
|
||||
if sh_prices:
|
||||
nearest_resistance[i] = sh_prices[-1]
|
||||
|
||||
c = close.iloc[i]
|
||||
if not np.isnan(nearest_support[i]) and not np.isnan(nearest_resistance[i]):
|
||||
zone_range = nearest_resistance[i] - nearest_support[i]
|
||||
if zone_range > 0:
|
||||
pos_pct = (c - nearest_support[i]) / zone_range
|
||||
in_demand_zone[i] = pos_pct < 0.35
|
||||
in_supply_zone[i] = pos_pct > 0.65
|
||||
|
||||
return DataFrame({
|
||||
"trend_up": trend_up_arr,
|
||||
"trend_down": trend_down_arr,
|
||||
"support": nearest_support,
|
||||
"resistance": nearest_resistance,
|
||||
"in_demand": in_demand_zone,
|
||||
"in_supply": in_supply_zone,
|
||||
}, index=high.index)
|
||||
|
||||
# =====================
|
||||
# 工具：K线形态检测
|
||||
# =====================
|
||||
|
||||
@staticmethod
|
||||
def _detect_candle_patterns(
|
||||
open_: pd.Series,
|
||||
high: pd.Series,
|
||||
low: pd.Series,
|
||||
close: pd.Series,
|
||||
pin_bar_wick_ratio: float = 0.6,
|
||||
) -> tuple[pd.Series, pd.Series, pd.Series, pd.Series]:
|
||||
body = (close - open_).abs()
|
||||
total_range = (high - low).replace(0, 0.0001)
|
||||
|
||||
upper_wick = high - close.where(close > open_, open_)
|
||||
lower_wick = open_.where(close > open_, close) - low
|
||||
is_pin = (upper_wick + lower_wick) / total_range > pin_bar_wick_ratio
|
||||
|
||||
bullish_pin = is_pin & (close > open_) & (lower_wick > upper_wick)
|
||||
bearish_pin = is_pin & (close < open_) & (upper_wick > lower_wick)
|
||||
|
||||
prev_open = open_.shift(1)
|
||||
prev_close = close.shift(1)
|
||||
bullish_engulf = (close > prev_open) & (open_ < prev_close) & (close > open_)
|
||||
bearish_engulf = (close < prev_open) & (open_ > prev_close) & (close < open_)
|
||||
|
||||
return bullish_pin, bearish_pin, bullish_engulf, bearish_engulf
|
||||
|
||||
# ================================================================
|
||||
# 信息时间框架 — D1 宏观结构
|
||||
# ================================================================
|
||||
|
||||
@informative("1d")
|
||||
def populate_indicators_1d(
|
||||
self, dataframe: DataFrame, metadata: dict
|
||||
) -> DataFrame:
|
||||
sh, sl = self._detect_swing_points(
|
||||
dataframe["high"], dataframe["low"],
|
||||
self.swing_lookback_d1.value,
|
||||
)
|
||||
structure = self._build_structure(
|
||||
dataframe["high"], dataframe["low"], dataframe["close"],
|
||||
sh, sl,
|
||||
)
|
||||
dataframe["trend_up"] = structure["trend_up"]
|
||||
dataframe["trend_down"] = structure["trend_down"]
|
||||
return dataframe
|
||||
|
||||
# ================================================================
|
||||
# 信息时间框架 — 4H 中期结构
|
||||
# ================================================================
|
||||
|
||||
@informative("4h")
|
||||
def populate_indicators_4h(
|
||||
self, dataframe: DataFrame, metadata: dict
|
||||
) -> DataFrame:
|
||||
sh, sl = self._detect_swing_points(
|
||||
dataframe["high"], dataframe["low"],
|
||||
self.swing_lookback_h4.value,
|
||||
)
|
||||
structure = self._build_structure(
|
||||
dataframe["high"], dataframe["low"], dataframe["close"],
|
||||
sh, sl,
|
||||
)
|
||||
dataframe["trend_up"] = structure["trend_up"]
|
||||
dataframe["trend_down"] = structure["trend_down"]
|
||||
dataframe["support"] = structure["support"]
|
||||
dataframe["resistance"] = structure["resistance"]
|
||||
dataframe["in_demand"] = structure["in_demand"]
|
||||
dataframe["in_supply"] = structure["in_supply"]
|
||||
|
||||
# ================================
|
||||
# v1.6 活支撑/阻力检查（保留）
|
||||
# ================================
|
||||
touched_support = (
|
||||
(dataframe["low"] <= dataframe["support"] * 1.005) &
|
||||
(dataframe["low"] >= dataframe["support"] * 0.995)
|
||||
)
|
||||
held_support = dataframe["close"] > dataframe["support"]
|
||||
support_tested_and_held = touched_support & held_support
|
||||
dataframe["support_alive"] = support_tested_and_held.rolling(3, min_periods=1).max() > 0
|
||||
|
||||
touched_resistance = (
|
||||
(dataframe["high"] >= dataframe["resistance"] * 0.995) &
|
||||
(dataframe["high"] <= dataframe["resistance"] * 1.005)
|
||||
)
|
||||
held_resistance = dataframe["close"] < dataframe["resistance"]
|
||||
resistance_tested_and_held = touched_resistance & held_resistance
|
||||
dataframe["resistance_alive"] = resistance_tested_and_held.rolling(3, min_periods=1).max() > 0
|
||||
|
||||
# ================================
|
||||
# v2.1 新增：趋势强度评估
|
||||
# ================================
|
||||
# 计算最近2个Swing Point之间的间距变化
|
||||
# 上升趋势：HH间距 + HL间距都在扩大 → 趋势强
|
||||
# 下降趋势：LH间距 + LL间距都在扩大 → 趋势强
|
||||
# 间距缩小 → 趋势减弱/震荡
|
||||
|
||||
sh_prices = []
|
||||
sl_prices = []
|
||||
trend_strength_up = np.full(len(dataframe), np.nan)
|
||||
trend_strength_down = np.full(len(dataframe), np.nan)
|
||||
|
||||
for i in range(len(dataframe)):
|
||||
if pd.notna(sh.iloc[i]):
|
||||
sh_prices.append(sh.iloc[i])
|
||||
if len(sh_prices) > 4:
|
||||
sh_prices.pop(0)
|
||||
if pd.notna(sl.iloc[i]):
|
||||
sl_prices.append(sl.iloc[i])
|
||||
if len(sl_prices) > 4:
|
||||
sl_prices.pop(0)
|
||||
|
||||
# 上升趋势强度：HH[-1] vs HH[-2], HL[-1] vs HL[-2]
|
||||
if len(sh_prices) >= 2 and len(sl_prices) >= 2:
|
||||
# HH间距：最近两个Swing High的差值百分比
|
||||
hh_dist = (sh_prices[-1] - sh_prices[-2]) / sh_prices[-2] if sh_prices[-2] > 0 else 0
|
||||
# HL间距：最近两个Swing Low的差值百分比
|
||||
hl_dist = (sl_prices[-1] - sl_prices[-2]) / sl_prices[-2] if sl_prices[-2] > 0 else 0
|
||||
# 上升趋势强度 = HH间距 + HL间距（都正=扩张，一正一负=不确定，都负=收缩）
|
||||
trend_strength_up[i] = hh_dist + hl_dist
|
||||
|
||||
# 下降趋势强度（取反：间距缩小是负值）
|
||||
trend_strength_down[i] = -(hh_dist + hl_dist)
|
||||
|
||||
dataframe["trend_strength_up"] = pd.Series(trend_strength_up, index=dataframe.index)
|
||||
dataframe["trend_strength_down"] = pd.Series(trend_strength_down, index=dataframe.index)
|
||||
|
||||
# 趋势强度是否足够（扩张中）
|
||||
min_strength = self.trend_strength_min.value / 100.0 # 0~0.30
|
||||
dataframe["strong_uptrend"] = dataframe["trend_strength_up"] > min_strength
|
||||
dataframe["strong_downtrend"] = dataframe["trend_strength_down"] > min_strength
|
||||
|
||||
return dataframe
|
||||
|
||||
# ================================================================
|
||||
# 主时间框架 — 1H 指标
|
||||
# ================================================================
|
||||
|
||||
def populate_indicators(
|
||||
self, dataframe: DataFrame, metadata: dict
|
||||
) -> DataFrame:
|
||||
"""1H 级别：K线形态（零指标）。"""
|
||||
bullish_pin, bearish_pin, bullish_engulf, bearish_engulf = (
|
||||
self._detect_candle_patterns(
|
||||
dataframe["open"],
|
||||
dataframe["high"],
|
||||
dataframe["low"],
|
||||
dataframe["close"],
|
||||
self.pin_bar_wick_ratio.value / 100.0,
|
||||
)
|
||||
)
|
||||
dataframe["bullish_pinbar"] = bullish_pin
|
||||
dataframe["bearish_pinbar"] = bearish_pin
|
||||
dataframe["bullish_engulfing"] = bullish_engulf
|
||||
dataframe["bearish_engulfing"] = bearish_engulf
|
||||
dataframe["bullish_signal"] = bullish_pin | bullish_engulf
|
||||
dataframe["bearish_signal"] = bearish_pin | bearish_engulf
|
||||
|
||||
# NaN 安全处理
|
||||
bool_cols = [
|
||||
"trend_up_1d", "trend_down_1d",
|
||||
"trend_up_4h", "trend_down_4h",
|
||||
"in_demand_4h", "in_supply_4h",
|
||||
"support_alive_4h", "resistance_alive_4h",
|
||||
"strong_uptrend_4h", "strong_downtrend_4h",
|
||||
"bullish_signal", "bearish_signal",
|
||||
]
|
||||
for col in bool_cols:
|
||||
if col in dataframe.columns:
|
||||
dataframe[col] = dataframe[col].fillna(False)
|
||||
|
||||
return dataframe
|
||||
|
||||
# =====================
|
||||
# 入场信号
|
||||
# =====================
|
||||
|
||||
def populate_entry_trend(self, dataframe: DataFrame, metadata: dict) -> DataFrame:
|
||||
"""
|
||||
入场逻辑（1H 时间框架）。
|
||||
|
||||
v2.1 核心改动：D1 — 趋势强度过滤
|
||||
做多额外条件：4H上升趋势在扩张（strong_uptrend_4h）
|
||||
做空额外条件：4H下降趋势在扩张（strong_downtrend_4h）
|
||||
"""
|
||||
max_dist = self.max_stop_dist.value / 100.0
|
||||
cooldown = self.cooldown_bars.value
|
||||
|
||||
# NaN 安全处理
|
||||
bool_cols = [
|
||||
"trend_up_1d", "trend_down_1d",
|
||||
"trend_up_4h", "trend_down_4h",
|
||||
"in_demand_4h", "in_supply_4h",
|
||||
"support_alive_4h", "resistance_alive_4h",
|
||||
"strong_uptrend_4h", "strong_downtrend_4h",
|
||||
"bullish_signal", "bearish_signal",
|
||||
]
|
||||
for col in bool_cols:
|
||||
if col in dataframe.columns:
|
||||
dataframe[col] = dataframe[col].fillna(False)
|
||||
|
||||
# ── 做多 ──
|
||||
long_stop_dist = (dataframe["open"] - dataframe["support_4h"]) / dataframe["open"]
|
||||
|
||||
long_base = (
|
||||
dataframe["trend_up_1d"]
|
||||
& dataframe["in_demand_4h"]
|
||||
& dataframe["bullish_signal"]
|
||||
& (long_stop_dist <= max_dist)
|
||||
& (long_stop_dist > 0.003)
|
||||
& dataframe["support_alive_4h"]
|
||||
# v2.1: 趋势强度 — 4H上升趋势必须在扩张
|
||||
& dataframe["strong_uptrend_4h"]
|
||||
)
|
||||
|
||||
long_recent = long_base.rolling(cooldown, min_periods=1).max().shift(1) == 0
|
||||
dataframe.loc[long_base & long_recent, "enter_long"] = 1
|
||||
|
||||
# ── 做空 ──
|
||||
short_stop_dist = (dataframe["resistance_4h"] - dataframe["open"]) / dataframe["open"]
|
||||
|
||||
short_base = (
|
||||
dataframe["trend_down_1d"]
|
||||
& dataframe["in_supply_4h"]
|
||||
& dataframe["bearish_signal"]
|
||||
& (short_stop_dist <= max_dist)
|
||||
& (short_stop_dist > 0.003)
|
||||
& dataframe["resistance_alive_4h"]
|
||||
# v2.1: 趋势强度 — 4H下降趋势必须在扩张
|
||||
& dataframe["strong_downtrend_4h"]
|
||||
)
|
||||
|
||||
short_recent = short_base.rolling(cooldown, min_periods=1).max().shift(1) == 0
|
||||
dataframe.loc[short_base & short_recent, "enter_short"] = 1
|
||||
|
||||
return dataframe
|
||||
|
||||
# =====================
|
||||
# 出场信号
|
||||
# =====================
|
||||
|
||||
def populate_exit_trend(self, dataframe: DataFrame, metadata: dict) -> DataFrame:
|
||||
"""出场逻辑 — 由结构反转触发。"""
|
||||
exit_long = ~dataframe["trend_up_1d"].fillna(True)
|
||||
dataframe.loc[exit_long, "exit_long"] = 1
|
||||
|
||||
exit_short = dataframe["trend_up_1d"].fillna(False)
|
||||
dataframe.loc[exit_short, "exit_short"] = 1
|
||||
|
||||
return dataframe
|
||||
|
||||
# =====================
|
||||
# 动态止损 — 纯价格结构（基于Swing Point）
|
||||
# =====================
|
||||
|
||||
def custom_stoploss(
|
||||
self,
|
||||
pair: str,
|
||||
trade: Trade,
|
||||
current_time: datetime,
|
||||
current_rate: float,
|
||||
current_profit: float,
|
||||
after_fill: bool,
|
||||
**kwargs,
|
||||
) -> float:
|
||||
"""
|
||||
止损逻辑：完全基于价格结构，零指标（与v1.6相同）。
|
||||
"""
|
||||
dataframe, _ = self.dp.get_analyzed_dataframe(pair, self.timeframe)
|
||||
if dataframe is None or len(dataframe) == 0:
|
||||
return -0.02 if not trade.is_short else 0.02
|
||||
|
||||
last = dataframe.iloc[-1]
|
||||
|
||||
if not trade.is_short:
|
||||
support = last.get("support_4h", np.nan)
|
||||
if pd.isna(support) or support <= 0:
|
||||
return -0.02
|
||||
sl_price = support * 0.999
|
||||
sl_ratio = (sl_price / current_rate) - 1.0
|
||||
return max(sl_ratio, -0.15)
|
||||
else:
|
||||
resistance = last.get("resistance_4h", np.nan)
|
||||
if pd.isna(resistance) or resistance <= 0:
|
||||
return 0.02
|
||||
sl_price = resistance * 1.001
|
||||
sl_ratio = 1.0 - (sl_price / current_rate)
|
||||
return min(sl_ratio, 0.15)
|
||||
|
||||
# =====================
|
||||
# Plot config
|
||||
# =====================
|
||||
|
||||
@staticmethod
|
||||
def plot_config() -> dict:
|
||||
return {
|
||||
"main_plot": {
|
||||
"support_4h": {"color": "green", "type": "line"},
|
||||
"resistance_4h": {"color": "red", "type": "line"},
|
||||
},
|
||||
"subplots": {
|
||||
"signals": {
|
||||
"bullish_pinbar": {"color": "green", "type": "scatter"},
|
||||
"bearish_pinbar": {"color": "red", "type": "scatter"},
|
||||
},
|
||||
"filters": {
|
||||
"support_alive_4h": {"color": "green", "type": "line"},
|
||||
"resistance_alive_4h": {"color": "red", "type": "line"},
|
||||
"strong_uptrend_4h": {"color": "blue", "type": "line"},
|
||||
"strong_downtrend_4h": {"color": "orange", "type": "line"},
|
||||
},
|
||||
},
|
||||
}
|
||||
@ -0,0 +1 @@
|
||||
{"max_open_trades":3,"stake_currency":"USDT","stake_amount":"unlimited","tradable_balance_ratio":0.99,"fiat_display_currency":"USD","dry_run":true,"dry_run_wallet":1000,"cancel_open_orders_on_exit":false,"trading_mode":"futures","margin_mode":"isolated","unfilledtimeout":{"entry":10,"exit":10,"exit_timeout_count":0,"unit":"minutes"},"entry_pricing":{"price_side":"same","use_order_book":true,"order_book_top":1,"price_last_balance":0.0,"check_depth_of_market":{"enabled":false,"bids_to_ask_delta":1}},"exit_pricing":{"price_side":"same","use_order_book":true,"order_book_top":1},"exchange":{"name":"binance","key":"REDACTED","secret":"REDACTED","ccxt_config":{"proxies":{"http":"http://host.docker.internal:7890","https":"http://host.docker.internal:7890"}},"ccxt_async_config":{"proxies":{"http":"http://host.docker.internal:7890","https":"http://host.docker.internal:7890"}},"pair_whitelist":["ETH/USDT:USDT"],"pair_blacklist":[]},"pairlists":[{"method":"StaticPairList"}],"bot_name":"freqtrade","initial_state":"running","internals":{"process_throttle_secs":5},"config_files":["/tmp/futures_config.json"]}
|
||||
Binary file not shown.
File diff suppressed because one or more lines are too long
@ -0,0 +1,456 @@
|
||||
"""
|
||||
Structure Flow Strategy v2.1
|
||||
=======================
|
||||
变更记录：
|
||||
v1.6 (2026-06-07): 最优基线 — +3659.63%, 190笔, 69.3% trailing胜率
|
||||
v2.0 (2026-06-08): B1 入场延迟确认 — 方向正确但降频严重
|
||||
v2.1 (2026-06-08): ===== D1: 趋势强度过滤 =====
|
||||
在4H级别评估趋势强度：最近2个Swing Point的间距变化。
|
||||
如果趋势在扩张（HH/HL间距增大），允许入场；
|
||||
如果趋势在收缩（HH/HL间距缩小或震荡），过滤信号。
|
||||
目的：只在趋势明确时交易，避免震荡市反复止损。
|
||||
"""
|
||||
|
||||
from datetime import datetime
|
||||
import numpy as np
|
||||
import pandas as pd
|
||||
from pandas import DataFrame
|
||||
from freqtrade.strategy import IStrategy, IntParameter, informative
|
||||
from freqtrade.persistence import Trade
|
||||
|
||||
|
||||
class StructureFlowStrategyV21(IStrategy):
|
||||
"""
|
||||
Structure Flow Strategy v2.1 — D1: 趋势强度过滤
|
||||
|
||||
v2.1改动（相对于v1.6）：
|
||||
在4H级别计算趋势强度：最近2个Swing High间距 + Swing Low间距的变化。
|
||||
只有趋势在扩张（或至少不收缩）时才允许入场。
|
||||
"""
|
||||
|
||||
can_short = True
|
||||
stoploss = -0.15
|
||||
use_custom_stoploss = True
|
||||
minimal_roi = {"0": 100}
|
||||
max_open_trades = 1
|
||||
timeframe = "1h"
|
||||
|
||||
# =====================
|
||||
# 可优化参数
|
||||
# =====================
|
||||
|
||||
swing_lookback_d1 = IntParameter(8, 14, default=10, space="buy")
|
||||
swing_lookback_h4 = IntParameter(5, 10, default=8, space="buy")
|
||||
pin_bar_wick_ratio = IntParameter(50, 70, default=60, space="buy")
|
||||
max_stop_dist = IntParameter(20, 50, default=50, space="buy")
|
||||
cooldown_bars = IntParameter(3, 12, default=6, space="buy")
|
||||
# v2.1 新增：趋势强度最小扩张比例（x/100 = 0%~50%）
|
||||
# 0 = 只要不收缩就行；越大要求趋势扩张越强
|
||||
trend_strength_min = IntParameter(-50, 20, default=-20, space="buy") # 扫描更宽范围
|
||||
|
||||
# =====================
|
||||
# 工具：Swing Point 检测
|
||||
# =====================
|
||||
|
||||
@staticmethod
|
||||
def _detect_swing_points(
|
||||
high: pd.Series,
|
||||
low: pd.Series,
|
||||
window: int = 5,
|
||||
) -> tuple[pd.Series, pd.Series]:
|
||||
n = len(high)
|
||||
sh = pd.Series(np.nan, index=high.index, dtype=float)
|
||||
sl = pd.Series(np.nan, index=low.index, dtype=float)
|
||||
|
||||
for i in range(window, n - window):
|
||||
if high.iloc[i] > high.iloc[i - window : i].max() and high.iloc[i] > high.iloc[i + 1 : i + window + 1].max():
|
||||
sh.iloc[i] = high.iloc[i]
|
||||
if low.iloc[i] < low.iloc[i - window : i].min() and low.iloc[i] < low.iloc[i + 1 : i + window + 1].min():
|
||||
sl.iloc[i] = low.iloc[i]
|
||||
|
||||
return sh, sl
|
||||
|
||||
# =====================
|
||||
# 工具：结构分析
|
||||
# =====================
|
||||
|
||||
def _build_structure(
|
||||
self,
|
||||
high: pd.Series,
|
||||
low: pd.Series,
|
||||
close: pd.Series,
|
||||
swing_high: pd.Series,
|
||||
swing_low: pd.Series,
|
||||
) -> DataFrame:
|
||||
n = len(high)
|
||||
|
||||
trend_up_arr = np.full(n, False)
|
||||
trend_down_arr = np.full(n, False)
|
||||
nearest_support = np.full(n, np.nan)
|
||||
nearest_resistance = np.full(n, np.nan)
|
||||
in_demand_zone = np.full(n, False)
|
||||
in_supply_zone = np.full(n, False)
|
||||
|
||||
sh_prices = []
|
||||
sl_prices = []
|
||||
|
||||
for i in range(n):
|
||||
if pd.notna(swing_high.iloc[i]):
|
||||
sh_prices.append(swing_high.iloc[i])
|
||||
if len(sh_prices) > 4:
|
||||
sh_prices.pop(0)
|
||||
|
||||
if pd.notna(swing_low.iloc[i]):
|
||||
sl_prices.append(swing_low.iloc[i])
|
||||
if len(sl_prices) > 4:
|
||||
sl_prices.pop(0)
|
||||
|
||||
if len(sh_prices) >= 2 and len(sl_prices) >= 2:
|
||||
if sh_prices[-1] > sh_prices[-2] and sl_prices[-1] > sl_prices[-2]:
|
||||
trend_up_arr[i] = True
|
||||
elif sh_prices[-1] < sh_prices[-2] and sl_prices[-1] < sl_prices[-2]:
|
||||
trend_down_arr[i] = True
|
||||
elif i > 0:
|
||||
trend_up_arr[i] = trend_up_arr[i - 1]
|
||||
trend_down_arr[i] = trend_down_arr[i - 1]
|
||||
elif i > 0:
|
||||
trend_up_arr[i] = trend_up_arr[i - 1]
|
||||
trend_down_arr[i] = trend_down_arr[i - 1]
|
||||
|
||||
if sl_prices:
|
||||
nearest_support[i] = sl_prices[-1]
|
||||
if sh_prices:
|
||||
nearest_resistance[i] = sh_prices[-1]
|
||||
|
||||
c = close.iloc[i]
|
||||
if not np.isnan(nearest_support[i]) and not np.isnan(nearest_resistance[i]):
|
||||
zone_range = nearest_resistance[i] - nearest_support[i]
|
||||
if zone_range > 0:
|
||||
pos_pct = (c - nearest_support[i]) / zone_range
|
||||
in_demand_zone[i] = pos_pct < 0.35
|
||||
in_supply_zone[i] = pos_pct > 0.65
|
||||
|
||||
return DataFrame({
|
||||
"trend_up": trend_up_arr,
|
||||
"trend_down": trend_down_arr,
|
||||
"support": nearest_support,
|
||||
"resistance": nearest_resistance,
|
||||
"in_demand": in_demand_zone,
|
||||
"in_supply": in_supply_zone,
|
||||
}, index=high.index)
|
||||
|
||||
# =====================
|
||||
# 工具：K线形态检测
|
||||
# =====================
|
||||
|
||||
@staticmethod
|
||||
def _detect_candle_patterns(
|
||||
open_: pd.Series,
|
||||
high: pd.Series,
|
||||
low: pd.Series,
|
||||
close: pd.Series,
|
||||
pin_bar_wick_ratio: float = 0.6,
|
||||
) -> tuple[pd.Series, pd.Series, pd.Series, pd.Series]:
|
||||
body = (close - open_).abs()
|
||||
total_range = (high - low).replace(0, 0.0001)
|
||||
|
||||
upper_wick = high - close.where(close > open_, open_)
|
||||
lower_wick = open_.where(close > open_, close) - low
|
||||
is_pin = (upper_wick + lower_wick) / total_range > pin_bar_wick_ratio
|
||||
|
||||
bullish_pin = is_pin & (close > open_) & (lower_wick > upper_wick)
|
||||
bearish_pin = is_pin & (close < open_) & (upper_wick > lower_wick)
|
||||
|
||||
prev_open = open_.shift(1)
|
||||
prev_close = close.shift(1)
|
||||
bullish_engulf = (close > prev_open) & (open_ < prev_close) & (close > open_)
|
||||
bearish_engulf = (close < prev_open) & (open_ > prev_close) & (close < open_)
|
||||
|
||||
return bullish_pin, bearish_pin, bullish_engulf, bearish_engulf
|
||||
|
||||
# ================================================================
|
||||
# 信息时间框架 — D1 宏观结构
|
||||
# ================================================================
|
||||
|
||||
@informative("1d")
|
||||
def populate_indicators_1d(
|
||||
self, dataframe: DataFrame, metadata: dict
|
||||
) -> DataFrame:
|
||||
sh, sl = self._detect_swing_points(
|
||||
dataframe["high"], dataframe["low"],
|
||||
self.swing_lookback_d1.value,
|
||||
)
|
||||
structure = self._build_structure(
|
||||
dataframe["high"], dataframe["low"], dataframe["close"],
|
||||
sh, sl,
|
||||
)
|
||||
dataframe["trend_up"] = structure["trend_up"]
|
||||
dataframe["trend_down"] = structure["trend_down"]
|
||||
return dataframe
|
||||
|
||||
# ================================================================
|
||||
# 信息时间框架 — 4H 中期结构
|
||||
# ================================================================
|
||||
|
||||
@informative("4h")
|
||||
def populate_indicators_4h(
|
||||
self, dataframe: DataFrame, metadata: dict
|
||||
) -> DataFrame:
|
||||
sh, sl = self._detect_swing_points(
|
||||
dataframe["high"], dataframe["low"],
|
||||
self.swing_lookback_h4.value,
|
||||
)
|
||||
structure = self._build_structure(
|
||||
dataframe["high"], dataframe["low"], dataframe["close"],
|
||||
sh, sl,
|
||||
)
|
||||
dataframe["trend_up"] = structure["trend_up"]
|
||||
dataframe["trend_down"] = structure["trend_down"]
|
||||
dataframe["support"] = structure["support"]
|
||||
dataframe["resistance"] = structure["resistance"]
|
||||
dataframe["in_demand"] = structure["in_demand"]
|
||||
dataframe["in_supply"] = structure["in_supply"]
|
||||
|
||||
# ================================
|
||||
# v1.6 活支撑/阻力检查（保留）
|
||||
# ================================
|
||||
touched_support = (
|
||||
(dataframe["low"] <= dataframe["support"] * 1.005) &
|
||||
(dataframe["low"] >= dataframe["support"] * 0.995)
|
||||
)
|
||||
held_support = dataframe["close"] > dataframe["support"]
|
||||
support_tested_and_held = touched_support & held_support
|
||||
dataframe["support_alive"] = support_tested_and_held.rolling(3, min_periods=1).max() > 0
|
||||
|
||||
touched_resistance = (
|
||||
(dataframe["high"] >= dataframe["resistance"] * 0.995) &
|
||||
(dataframe["high"] <= dataframe["resistance"] * 1.005)
|
||||
)
|
||||
held_resistance = dataframe["close"] < dataframe["resistance"]
|
||||
resistance_tested_and_held = touched_resistance & held_resistance
|
||||
dataframe["resistance_alive"] = resistance_tested_and_held.rolling(3, min_periods=1).max() > 0
|
||||
|
||||
# ================================
|
||||
# v2.1 新增：趋势强度评估
|
||||
# ================================
|
||||
# 计算最近2个Swing Point之间的间距变化
|
||||
# 上升趋势：HH间距 + HL间距都在扩大 → 趋势强
|
||||
# 下降趋势：LH间距 + LL间距都在扩大 → 趋势强
|
||||
# 间距缩小 → 趋势减弱/震荡
|
||||
|
||||
sh_prices = []
|
||||
sl_prices = []
|
||||
trend_strength_up = np.full(len(dataframe), np.nan)
|
||||
trend_strength_down = np.full(len(dataframe), np.nan)
|
||||
|
||||
for i in range(len(dataframe)):
|
||||
if pd.notna(sh.iloc[i]):
|
||||
sh_prices.append(sh.iloc[i])
|
||||
if len(sh_prices) > 4:
|
||||
sh_prices.pop(0)
|
||||
if pd.notna(sl.iloc[i]):
|
||||
sl_prices.append(sl.iloc[i])
|
||||
if len(sl_prices) > 4:
|
||||
sl_prices.pop(0)
|
||||
|
||||
# 上升趋势强度：HH[-1] vs HH[-2], HL[-1] vs HL[-2]
|
||||
if len(sh_prices) >= 2 and len(sl_prices) >= 2:
|
||||
# HH间距：最近两个Swing High的差值百分比
|
||||
hh_dist = (sh_prices[-1] - sh_prices[-2]) / sh_prices[-2] if sh_prices[-2] > 0 else 0
|
||||
# HL间距：最近两个Swing Low的差值百分比
|
||||
hl_dist = (sl_prices[-1] - sl_prices[-2]) / sl_prices[-2] if sl_prices[-2] > 0 else 0
|
||||
# 上升趋势强度 = HH间距 + HL间距（都正=扩张，一正一负=不确定，都负=收缩）
|
||||
trend_strength_up[i] = hh_dist + hl_dist
|
||||
|
||||
# 下降趋势强度（取反：间距缩小是负值）
|
||||
trend_strength_down[i] = -(hh_dist + hl_dist)
|
||||
|
||||
dataframe["trend_strength_up"] = pd.Series(trend_strength_up, index=dataframe.index)
|
||||
dataframe["trend_strength_down"] = pd.Series(trend_strength_down, index=dataframe.index)
|
||||
|
||||
# 趋势强度是否足够（扩张中）
|
||||
min_strength = self.trend_strength_min.value / 100.0 # 0~0.30
|
||||
dataframe["strong_uptrend"] = dataframe["trend_strength_up"] > min_strength
|
||||
dataframe["strong_downtrend"] = dataframe["trend_strength_down"] > min_strength
|
||||
|
||||
return dataframe
|
||||
|
||||
# ================================================================
|
||||
# 主时间框架 — 1H 指标
|
||||
# ================================================================
|
||||
|
||||
def populate_indicators(
|
||||
self, dataframe: DataFrame, metadata: dict
|
||||
) -> DataFrame:
|
||||
"""1H 级别：K线形态（零指标）。"""
|
||||
bullish_pin, bearish_pin, bullish_engulf, bearish_engulf = (
|
||||
self._detect_candle_patterns(
|
||||
dataframe["open"],
|
||||
dataframe["high"],
|
||||
dataframe["low"],
|
||||
dataframe["close"],
|
||||
self.pin_bar_wick_ratio.value / 100.0,
|
||||
)
|
||||
)
|
||||
dataframe["bullish_pinbar"] = bullish_pin
|
||||
dataframe["bearish_pinbar"] = bearish_pin
|
||||
dataframe["bullish_engulfing"] = bullish_engulf
|
||||
dataframe["bearish_engulfing"] = bearish_engulf
|
||||
dataframe["bullish_signal"] = bullish_pin | bullish_engulf
|
||||
dataframe["bearish_signal"] = bearish_pin | bearish_engulf
|
||||
|
||||
# NaN 安全处理
|
||||
bool_cols = [
|
||||
"trend_up_1d", "trend_down_1d",
|
||||
"trend_up_4h", "trend_down_4h",
|
||||
"in_demand_4h", "in_supply_4h",
|
||||
"support_alive_4h", "resistance_alive_4h",
|
||||
"strong_uptrend_4h", "strong_downtrend_4h",
|
||||
"bullish_signal", "bearish_signal",
|
||||
]
|
||||
for col in bool_cols:
|
||||
if col in dataframe.columns:
|
||||
dataframe[col] = dataframe[col].fillna(False)
|
||||
|
||||
return dataframe
|
||||
|
||||
# =====================
|
||||
# 入场信号
|
||||
# =====================
|
||||
|
||||
def populate_entry_trend(self, dataframe: DataFrame, metadata: dict) -> DataFrame:
|
||||
"""
|
||||
入场逻辑（1H 时间框架）。
|
||||
|
||||
v2.1 核心改动：D1 — 趋势强度过滤
|
||||
做多额外条件：4H上升趋势在扩张（strong_uptrend_4h）
|
||||
做空额外条件：4H下降趋势在扩张（strong_downtrend_4h）
|
||||
"""
|
||||
max_dist = self.max_stop_dist.value / 100.0
|
||||
cooldown = self.cooldown_bars.value
|
||||
|
||||
# NaN 安全处理
|
||||
bool_cols = [
|
||||
"trend_up_1d", "trend_down_1d",
|
||||
"trend_up_4h", "trend_down_4h",
|
||||
"in_demand_4h", "in_supply_4h",
|
||||
"support_alive_4h", "resistance_alive_4h",
|
||||
"strong_uptrend_4h", "strong_downtrend_4h",
|
||||
"bullish_signal", "bearish_signal",
|
||||
]
|
||||
for col in bool_cols:
|
||||
if col in dataframe.columns:
|
||||
dataframe[col] = dataframe[col].fillna(False)
|
||||
|
||||
# ── 做多 ──
|
||||
long_stop_dist = (dataframe["open"] - dataframe["support_4h"]) / dataframe["open"]
|
||||
|
||||
long_base = (
|
||||
dataframe["trend_up_1d"]
|
||||
& dataframe["in_demand_4h"]
|
||||
& dataframe["bullish_signal"]
|
||||
& (long_stop_dist <= max_dist)
|
||||
& (long_stop_dist > 0.003)
|
||||
& dataframe["support_alive_4h"]
|
||||
# v2.1: 趋势强度 — 4H上升趋势必须在扩张
|
||||
& dataframe["strong_uptrend_4h"]
|
||||
)
|
||||
|
||||
long_recent = long_base.rolling(cooldown, min_periods=1).max().shift(1) == 0
|
||||
dataframe.loc[long_base & long_recent, "enter_long"] = 1
|
||||
|
||||
# ── 做空 ──
|
||||
short_stop_dist = (dataframe["resistance_4h"] - dataframe["open"]) / dataframe["open"]
|
||||
|
||||
short_base = (
|
||||
dataframe["trend_down_1d"]
|
||||
& dataframe["in_supply_4h"]
|
||||
& dataframe["bearish_signal"]
|
||||
& (short_stop_dist <= max_dist)
|
||||
& (short_stop_dist > 0.003)
|
||||
& dataframe["resistance_alive_4h"]
|
||||
# v2.1: 趋势强度 — 4H下降趋势必须在扩张
|
||||
& dataframe["strong_downtrend_4h"]
|
||||
)
|
||||
|
||||
short_recent = short_base.rolling(cooldown, min_periods=1).max().shift(1) == 0
|
||||
dataframe.loc[short_base & short_recent, "enter_short"] = 1
|
||||
|
||||
return dataframe
|
||||
|
||||
# =====================
|
||||
# 出场信号
|
||||
# =====================
|
||||
|
||||
def populate_exit_trend(self, dataframe: DataFrame, metadata: dict) -> DataFrame:
|
||||
"""出场逻辑 — 由结构反转触发。"""
|
||||
exit_long = ~dataframe["trend_up_1d"].fillna(True)
|
||||
dataframe.loc[exit_long, "exit_long"] = 1
|
||||
|
||||
exit_short = dataframe["trend_up_1d"].fillna(False)
|
||||
dataframe.loc[exit_short, "exit_short"] = 1
|
||||
|
||||
return dataframe
|
||||
|
||||
# =====================
|
||||
# 动态止损 — 纯价格结构（基于Swing Point）
|
||||
# =====================
|
||||
|
||||
def custom_stoploss(
|
||||
self,
|
||||
pair: str,
|
||||
trade: Trade,
|
||||
current_time: datetime,
|
||||
current_rate: float,
|
||||
current_profit: float,
|
||||
after_fill: bool,
|
||||
**kwargs,
|
||||
) -> float:
|
||||
"""
|
||||
止损逻辑：完全基于价格结构，零指标（与v1.6相同）。
|
||||
"""
|
||||
dataframe, _ = self.dp.get_analyzed_dataframe(pair, self.timeframe)
|
||||
if dataframe is None or len(dataframe) == 0:
|
||||
return -0.02 if not trade.is_short else 0.02
|
||||
|
||||
last = dataframe.iloc[-1]
|
||||
|
||||
if not trade.is_short:
|
||||
support = last.get("support_4h", np.nan)
|
||||
if pd.isna(support) or support <= 0:
|
||||
return -0.02
|
||||
sl_price = support * 0.999
|
||||
sl_ratio = (sl_price / current_rate) - 1.0
|
||||
return max(sl_ratio, -0.15)
|
||||
else:
|
||||
resistance = last.get("resistance_4h", np.nan)
|
||||
if pd.isna(resistance) or resistance <= 0:
|
||||
return 0.02
|
||||
sl_price = resistance * 1.001
|
||||
sl_ratio = 1.0 - (sl_price / current_rate)
|
||||
return min(sl_ratio, 0.15)
|
||||
|
||||
# =====================
|
||||
# Plot config
|
||||
# =====================
|
||||
|
||||
@staticmethod
|
||||
def plot_config() -> dict:
|
||||
return {
|
||||
"main_plot": {
|
||||
"support_4h": {"color": "green", "type": "line"},
|
||||
"resistance_4h": {"color": "red", "type": "line"},
|
||||
},
|
||||
"subplots": {
|
||||
"signals": {
|
||||
"bullish_pinbar": {"color": "green", "type": "scatter"},
|
||||
"bearish_pinbar": {"color": "red", "type": "scatter"},
|
||||
},
|
||||
"filters": {
|
||||
"support_alive_4h": {"color": "green", "type": "line"},
|
||||
"resistance_alive_4h": {"color": "red", "type": "line"},
|
||||
"strong_uptrend_4h": {"color": "blue", "type": "line"},
|
||||
"strong_downtrend_4h": {"color": "orange", "type": "line"},
|
||||
},
|
||||
},
|
||||
}
|
||||
@ -0,0 +1 @@
|
||||
{"max_open_trades":3,"stake_currency":"USDT","stake_amount":"unlimited","tradable_balance_ratio":0.99,"fiat_display_currency":"USD","dry_run":true,"dry_run_wallet":1000,"cancel_open_orders_on_exit":false,"trading_mode":"futures","margin_mode":"isolated","unfilledtimeout":{"entry":10,"exit":10,"exit_timeout_count":0,"unit":"minutes"},"entry_pricing":{"price_side":"same","use_order_book":true,"order_book_top":1,"price_last_balance":0.0,"check_depth_of_market":{"enabled":false,"bids_to_ask_delta":1}},"exit_pricing":{"price_side":"same","use_order_book":true,"order_book_top":1},"exchange":{"name":"binance","key":"REDACTED","secret":"REDACTED","ccxt_config":{"proxies":{"http":"http://host.docker.internal:7890","https":"http://host.docker.internal:7890"}},"ccxt_async_config":{"proxies":{"http":"http://host.docker.internal:7890","https":"http://host.docker.internal:7890"}},"pair_whitelist":["ETH/USDT:USDT"],"pair_blacklist":[]},"pairlists":[{"method":"StaticPairList"}],"bot_name":"freqtrade","initial_state":"running","internals":{"process_throttle_secs":5},"config_files":["/tmp/futures_config.json"]}
|
||||
Binary file not shown.
2288
backtest/v2_2c_full_2021_2026.txt
Normal file
2288
backtest/v2_2c_full_2021_2026.txt
Normal file
File diff suppressed because it is too large
Load Diff
2926
backtest/v2_2d_full_2021_2026.txt
Normal file
2926
backtest/v2_2d_full_2021_2026.txt
Normal file
File diff suppressed because it is too large
Load Diff
32
backtest/v2_2d_yearly_summary.md
Normal file
32
backtest/v2_2d_yearly_summary.md
Normal file
@ -0,0 +1,32 @@
|
||||
# v2.2d 逐年回测结果
|
||||
|
||||
**回测时间**: 2026-06-11 13:08
|
||||
**策略**: StructureFlowStrategyV22d
|
||||
**配置**: Docker freqtrade, config_backtest.json, ETH/USDT:USDT, futures, 1x, $10,000起
|
||||
|
||||
## 逐年表现（独立运行，每年 $10,000 起）
|
||||
|
||||
| 年份 | 交易数 | 收益率 | 终值 | 市场涨跌 | 胜率 | 最大回撤 |
|
||||
|------|--------|--------|------|----------|------|----------|
|
||||
| 2021 | 172 | +251.16% | $35,116 | +406.63% | 27.3% | 11.28% |
|
||||
| 2022 | 204 | +110.91% | $21,091 | -67.92% | 30.9% | 11.69% |
|
||||
| 2023 | 182 | +49.35% | $14,935 | +92.43% | 26.9% | 10.04% |
|
||||
| 2024 | 232 | +185.84% | $28,584 | +46.38% | 28.4% | 6.87% |
|
||||
| 2025 | 221 | +608.24% | $70,824 | -11.43% | 27.6% | 13.92% |
|
||||
| 2026 | 54 | -11.87% | $8,813 | -45.37% | 22.2% | 14.89% |
|
||||
|
||||
**逐年合计**: 1,065笔（独立运行，不跨年复合）
|
||||
|
||||
## 全周期（2021-2026 连续运行）
|
||||
|
||||
| 交易数 | 总收益率 | 终值 | CAGR | Sharpe | 最大回撤 |
|
||||
|--------|----------|------|------|--------|----------|
|
||||
| 1,375 | +205,684.36% | $20,578,436 | 309.01% | 1.03 | 20.58% |
|
||||
|
||||
## 关键观察
|
||||
|
||||
1. **熊市表现优异**: 2022市场-68%，策略+111%；2026YTD市场-45%，策略仅-12%
|
||||
2. **牛市相对逊色**: 2021市场+407%，策略+251%；2023市场+92%，策略+49%
|
||||
3. **2025是爆发年**: +608%，主因2025年ETH波动大（先跌后涨），双向策略充分获利
|
||||
4. **逐年回撤控制在7-15%**，比全周期20.58%低，因逐年重置不积累
|
||||
5. **逐年笔数稳定**: 172-232笔/年（2026仅半年54笔），日均0.5-0.6笔
|
||||
385
dashboard/indicators.py
Normal file
385
dashboard/indicators.py
Normal file
@ -0,0 +1,385 @@
|
||||
"""
|
||||
指标计算模块 — 从 v2.2b 策略提取的独立版本
|
||||
用于 APP 后端实时计算市场结构和信号诊断
|
||||
"""
|
||||
import numpy as np
|
||||
import pandas as pd
|
||||
|
||||
|
||||
def detect_swing_points(
|
||||
high: pd.Series,
|
||||
low: pd.Series,
|
||||
window: int = 5,
|
||||
) -> tuple[pd.Series, pd.Series]:
|
||||
"""检测 Swing High / Swing Low，与 v2.2b 策略一致"""
|
||||
n = len(high)
|
||||
sh = pd.Series(np.nan, index=high.index, dtype=float)
|
||||
sl = pd.Series(np.nan, index=low.index, dtype=float)
|
||||
|
||||
for i in range(window, n - window):
|
||||
if high.iloc[i] > high.iloc[i - window : i].max() and high.iloc[i] > high.iloc[i + 1 : i + window + 1].max():
|
||||
sh.iloc[i] = high.iloc[i]
|
||||
if low.iloc[i] < low.iloc[i - window : i].min() and low.iloc[i] < low.iloc[i + 1 : i + window + 1].min():
|
||||
sl.iloc[i] = low.iloc[i]
|
||||
|
||||
return sh, sl
|
||||
|
||||
|
||||
def build_structure(
|
||||
high: pd.Series,
|
||||
low: pd.Series,
|
||||
close: pd.Series,
|
||||
swing_high: pd.Series,
|
||||
swing_low: pd.Series,
|
||||
) -> pd.DataFrame:
|
||||
"""构建价格结构（趋势方向、S/R 位、供需区），与 v2.2b 一致"""
|
||||
n = len(high)
|
||||
|
||||
trend_up_arr = np.full(n, False)
|
||||
trend_down_arr = np.full(n, False)
|
||||
nearest_support = np.full(n, np.nan)
|
||||
nearest_resistance = np.full(n, np.nan)
|
||||
in_demand_zone = np.full(n, False)
|
||||
in_supply_zone = np.full(n, False)
|
||||
|
||||
sh_prices = []
|
||||
sl_prices = []
|
||||
|
||||
for i in range(n):
|
||||
if pd.notna(swing_high.iloc[i]):
|
||||
sh_prices.append(swing_high.iloc[i])
|
||||
if len(sh_prices) > 4:
|
||||
sh_prices.pop(0)
|
||||
|
||||
if pd.notna(swing_low.iloc[i]):
|
||||
sl_prices.append(swing_low.iloc[i])
|
||||
if len(sl_prices) > 4:
|
||||
sl_prices.pop(0)
|
||||
|
||||
if len(sh_prices) >= 2 and len(sl_prices) >= 2:
|
||||
if sh_prices[-1] > sh_prices[-2] and sl_prices[-1] > sl_prices[-2]:
|
||||
trend_up_arr[i] = True
|
||||
elif sh_prices[-1] < sh_prices[-2] and sl_prices[-1] < sl_prices[-2]:
|
||||
trend_down_arr[i] = True
|
||||
elif i > 0:
|
||||
trend_up_arr[i] = trend_up_arr[i - 1]
|
||||
trend_down_arr[i] = trend_down_arr[i - 1]
|
||||
elif i > 0:
|
||||
trend_up_arr[i] = trend_up_arr[i - 1]
|
||||
trend_down_arr[i] = trend_down_arr[i - 1]
|
||||
|
||||
if sl_prices:
|
||||
nearest_support[i] = sl_prices[-1]
|
||||
if sh_prices:
|
||||
nearest_resistance[i] = sh_prices[-1]
|
||||
|
||||
c = close.iloc[i]
|
||||
if not np.isnan(nearest_support[i]) and not np.isnan(nearest_resistance[i]):
|
||||
zone_range = nearest_resistance[i] - nearest_support[i]
|
||||
if zone_range > 0:
|
||||
pos_pct = (c - nearest_support[i]) / zone_range
|
||||
in_demand_zone[i] = pos_pct < 0.35
|
||||
in_supply_zone[i] = pos_pct > 0.65
|
||||
|
||||
return pd.DataFrame({
|
||||
"trend_up": trend_up_arr,
|
||||
"trend_down": trend_down_arr,
|
||||
"support": nearest_support,
|
||||
"resistance": nearest_resistance,
|
||||
"in_demand": in_demand_zone,
|
||||
"in_supply": in_supply_zone,
|
||||
}, index=high.index)
|
||||
|
||||
|
||||
def compute_trend_strength(
|
||||
high: pd.Series,
|
||||
low: pd.Series,
|
||||
swing_high: pd.Series,
|
||||
swing_low: pd.Series,
|
||||
min_strength: float = -0.20,
|
||||
) -> pd.DataFrame:
|
||||
"""计算趋势强度（4H Swing Point 间距变化），与 v2.2b 一致"""
|
||||
sh_prices = []
|
||||
sl_prices = []
|
||||
trend_strength_up = np.full(len(high), np.nan)
|
||||
trend_strength_down = np.full(len(high), np.nan)
|
||||
strong_uptrend = np.full(len(high), False)
|
||||
strong_downtrend = np.full(len(high), False)
|
||||
|
||||
for i in range(len(high)):
|
||||
if pd.notna(swing_high.iloc[i]):
|
||||
sh_prices.append(swing_high.iloc[i])
|
||||
if len(sh_prices) > 4:
|
||||
sh_prices.pop(0)
|
||||
if pd.notna(swing_low.iloc[i]):
|
||||
sl_prices.append(swing_low.iloc[i])
|
||||
if len(sl_prices) > 4:
|
||||
sl_prices.pop(0)
|
||||
|
||||
if len(sh_prices) >= 2 and len(sl_prices) >= 2:
|
||||
hh_dist = (sh_prices[-1] - sh_prices[-2]) / sh_prices[-2] if sh_prices[-2] > 0 else 0
|
||||
hl_dist = (sl_prices[-1] - sl_prices[-2]) / sl_prices[-2] if sl_prices[-2] > 0 else 0
|
||||
trend_strength_up[i] = hh_dist + hl_dist
|
||||
trend_strength_down[i] = -(hh_dist + hl_dist)
|
||||
|
||||
strong_uptrend = pd.Series(trend_strength_up) > min_strength
|
||||
strong_downtrend = pd.Series(trend_strength_down) > min_strength
|
||||
|
||||
return pd.DataFrame({
|
||||
"trend_strength_up": pd.Series(trend_strength_up, index=high.index),
|
||||
"trend_strength_down": pd.Series(trend_strength_down, index=high.index),
|
||||
"strong_uptrend": strong_uptrend.values,
|
||||
"strong_downtrend": strong_downtrend.values,
|
||||
}, index=high.index)
|
||||
|
||||
|
||||
def compute_support_alive(
|
||||
low: pd.Series,
|
||||
close: pd.Series,
|
||||
support: pd.Series,
|
||||
) -> pd.Series:
|
||||
"""检查支撑是否活着（最近测试过没跌破），与 v2.2b 一致"""
|
||||
touched = (
|
||||
(low <= support * 1.005) &
|
||||
(low >= support * 0.995)
|
||||
)
|
||||
held = close > support
|
||||
tested_and_held = touched & held
|
||||
return tested_and_held.rolling(3, min_periods=1).max() > 0
|
||||
|
||||
|
||||
def compute_resistance_alive(
|
||||
high: pd.Series,
|
||||
close: pd.Series,
|
||||
resistance: pd.Series,
|
||||
) -> pd.Series:
|
||||
"""检查阻力是否活着（最近测试过没突破），与 v2.2b 一致"""
|
||||
touched = (
|
||||
(high >= resistance * 0.995) &
|
||||
(high <= resistance * 1.005)
|
||||
)
|
||||
held = close < resistance
|
||||
tested_and_held = touched & held
|
||||
return tested_and_held.rolling(3, min_periods=1).max() > 0
|
||||
|
||||
|
||||
def detect_candle_patterns(
|
||||
open_: pd.Series,
|
||||
high: pd.Series,
|
||||
low: pd.Series,
|
||||
close: pd.Series,
|
||||
pin_bar_wick_ratio: float = 0.6,
|
||||
) -> dict:
|
||||
"""检测 K 线形态，与 v2.2b 一致"""
|
||||
body = (close - open_).abs()
|
||||
total_range = (high - low).replace(0, 0.0001)
|
||||
|
||||
upper_wick = high - close.where(close > open_, open_)
|
||||
lower_wick = open_.where(close > open_, close) - low
|
||||
is_pin = (upper_wick + lower_wick) / total_range > pin_bar_wick_ratio
|
||||
|
||||
bullish_pin = is_pin & (close > open_) & (lower_wick > upper_wick)
|
||||
bearish_pin = is_pin & (close < open_) & (upper_wick > lower_wick)
|
||||
|
||||
prev_open = open_.shift(1)
|
||||
prev_close = close.shift(1)
|
||||
bullish_engulf = (close > prev_open) & (open_ < prev_close) & (close > open_)
|
||||
bearish_engulf = (close < prev_open) & (open_ > prev_close) & (close < open_)
|
||||
|
||||
return {
|
||||
"bullish_pinbar": bullish_pin,
|
||||
"bearish_pinbar": bearish_pin,
|
||||
"bullish_signal": bullish_pin | bullish_engulf,
|
||||
"bearish_signal": bearish_pin | bearish_engulf,
|
||||
}
|
||||
|
||||
|
||||
def compute_all_indicators(
|
||||
df_1h: pd.DataFrame,
|
||||
df_4h: pd.DataFrame,
|
||||
df_1d: pd.DataFrame,
|
||||
params: dict = None,
|
||||
) -> dict:
|
||||
"""
|
||||
完整计算所有指标的入口函数
|
||||
返回 APP 需要的全部数据
|
||||
|
||||
params 可选参数：
|
||||
swing_lookback_d1: int (default 5)
|
||||
swing_lookback_h4: int (default 8)
|
||||
pin_bar_wick_ratio: float (default 0.6)
|
||||
max_stop_dist: float (default 0.50 = 5%)
|
||||
trend_strength_min: float (default -0.20)
|
||||
"""
|
||||
if params is None:
|
||||
params = {}
|
||||
|
||||
swing_lookback_d1 = params.get("swing_lookback_d1", 5)
|
||||
swing_lookback_h4 = params.get("swing_lookback_h4", 8)
|
||||
swing_lookback_1h = params.get("swing_lookback_1h", 5)
|
||||
pin_bar_wick_ratio = params.get("pin_bar_wick_ratio", 0.6)
|
||||
max_stop_dist = params.get("max_stop_dist", 0.50)
|
||||
trend_strength_min = params.get("trend_strength_min", -0.20)
|
||||
|
||||
# ---- D1 结构 ----
|
||||
sh_d1, sl_d1 = detect_swing_points(df_1d["high"], df_1d["low"], swing_lookback_d1)
|
||||
struct_d1 = build_structure(df_1d["high"], df_1d["low"], df_1d["close"], sh_d1, sl_d1)
|
||||
trend_up_1d = bool(struct_d1["trend_up"].iloc[-1]) if len(struct_d1) > 0 else False
|
||||
trend_down_1d = bool(struct_d1["trend_down"].iloc[-1]) if len(struct_d1) > 0 else False
|
||||
|
||||
# ---- 4H 结构（仅趋势强度） ----
|
||||
sh_4h, sl_4h = detect_swing_points(df_4h["high"], df_4h["low"], swing_lookback_h4)
|
||||
struct_4h = build_structure(df_4h["high"], df_4h["low"], df_4h["close"], sh_4h, sl_4h)
|
||||
strength_4h = compute_trend_strength(
|
||||
df_4h["high"], df_4h["low"], sh_4h, sl_4h,
|
||||
min_strength=trend_strength_min,
|
||||
)
|
||||
|
||||
strong_uptrend_4h = bool(strength_4h["strong_uptrend"].iloc[-1]) if len(strength_4h) > 0 else False
|
||||
strong_downtrend_4h = bool(strength_4h["strong_downtrend"].iloc[-1]) if len(strength_4h) > 0 else False
|
||||
|
||||
trend_strength_up_val = float(strength_4h["trend_strength_up"].iloc[-1]) if len(strength_4h) > 0 else 0
|
||||
trend_strength_down_val = float(strength_4h["trend_strength_down"].iloc[-1]) if len(strength_4h) > 0 else 0
|
||||
|
||||
# ---- 1H 结构（S/R 适配 v2.2c） ----
|
||||
sh_1h, sl_1h = detect_swing_points(df_1h["high"], df_1h["low"], swing_lookback_1h)
|
||||
struct_1h = build_structure(df_1h["high"], df_1h["low"], df_1h["close"], sh_1h, sl_1h)
|
||||
|
||||
last_1h_s = struct_1h.iloc[-1] if len(struct_1h) > 0 else None
|
||||
support_1h = float(last_1h_s["support"]) if last_1h_s is not None and pd.notna(last_1h_s["support"]) else None
|
||||
resistance_1h = float(last_1h_s["resistance"]) if last_1h_s is not None and pd.notna(last_1h_s["resistance"]) else None
|
||||
in_demand_1h = bool(last_1h_s["in_demand"]) if last_1h_s is not None else False
|
||||
in_supply_1h = bool(last_1h_s["in_supply"]) if last_1h_s is not None else False
|
||||
|
||||
# 1H 活 S/R 检查
|
||||
support_alive_1h = False
|
||||
resistance_alive_1h = False
|
||||
if support_1h is not None:
|
||||
support_alive_1h = bool(compute_support_alive(df_1h["low"], df_1h["close"], struct_1h["support"]).iloc[-1])
|
||||
if resistance_1h is not None:
|
||||
resistance_alive_1h = bool(compute_resistance_alive(df_1h["high"], df_1h["close"], struct_1h["resistance"]).iloc[-1])
|
||||
|
||||
# ---- 1H K线形态 ----
|
||||
last_1h_candle = df_1h.iloc[-1] if len(df_1h) > 0 else None
|
||||
candle = detect_candle_patterns(
|
||||
df_1h["open"], df_1h["high"], df_1h["low"], df_1h["close"],
|
||||
pin_bar_wick_ratio,
|
||||
)
|
||||
bullish_signal = bool(candle["bullish_signal"].iloc[-1]) if len(df_1h) > 0 else False
|
||||
bearish_signal = bool(candle["bearish_signal"].iloc[-1]) if len(df_1h) > 0 else False
|
||||
bullish_pinbar = bool(candle["bullish_pinbar"].iloc[-1]) if len(df_1h) > 0 else False
|
||||
bearish_pinbar = bool(candle["bearish_pinbar"].iloc[-1]) if len(df_1h) > 0 else False
|
||||
|
||||
# ---- 入场距离（1H S/R） ----
|
||||
current_price = float(df_1h["open"].iloc[-1]) if len(df_1h) > 0 else 0
|
||||
long_stop_dist = None
|
||||
short_stop_dist = None
|
||||
if support_1h is not None and support_1h > 0:
|
||||
long_stop_dist = abs((current_price - support_1h) / current_price)
|
||||
if resistance_1h is not None and resistance_1h > 0:
|
||||
short_stop_dist = abs((resistance_1h - current_price) / current_price)
|
||||
|
||||
long_dist_ok = long_stop_dist is not None and long_stop_dist <= max_stop_dist and long_stop_dist > 0.003
|
||||
short_dist_ok = short_stop_dist is not None and short_stop_dist <= max_stop_dist and short_stop_dist > 0.003
|
||||
|
||||
# ---- Swing Point 历史（1H 用于画图） ----
|
||||
swing_highs_1h = df_1h[sh_1h.notna()].index.strftime("%Y-%m-%d %H:%M").tolist() if len(sh_1h) > 0 else []
|
||||
swing_high_prices_1h = [float(x) for x in sh_1h.dropna().tolist()]
|
||||
swing_lows_1h = df_1h[sl_1h.notna()].index.strftime("%Y-%m-%d %H:%M").tolist() if len(sl_1h) > 0 else []
|
||||
swing_low_prices_1h = [float(x) for x in sl_1h.dropna().tolist()]
|
||||
|
||||
# ---- 信号诊断结果（1H S/R） ----
|
||||
diagnosis = {
|
||||
"market_price": round(current_price, 2),
|
||||
"support": round(support_1h, 2) if support_1h else None,
|
||||
"resistance": round(resistance_1h, 2) if resistance_1h else None,
|
||||
"zone_width_pct": round(abs(resistance_1h - support_1h) / support_1h * 100, 2) if support_1h and resistance_1h else None,
|
||||
"price_position_in_zone": round((current_price - support_1h) / (resistance_1h - support_1h) * 100, 1) if support_1h and resistance_1h and resistance_1h > support_1h else None,
|
||||
}
|
||||
|
||||
filters = {
|
||||
"trend_up_1d": {
|
||||
"pass": trend_up_1d,
|
||||
"desc": "D1 上升趋势（HH+HL 都在抬高）",
|
||||
"value": "上升" if trend_up_1d else ("下降" if trend_down_1d else "震荡"),
|
||||
},
|
||||
"in_demand_1h": {
|
||||
"pass": in_demand_1h,
|
||||
"desc": "价格在 1H 需求区（zone 底部 35%）",
|
||||
"value": f"位置 {diagnosis['price_position_in_zone']}%" if diagnosis['price_position_in_zone'] is not None else "N/A",
|
||||
},
|
||||
"long_stop_distance": {
|
||||
"pass": long_dist_ok,
|
||||
"desc": f"入场距支撑距离 ({'%.2f' % (long_stop_dist*100) if long_stop_dist else 'N/A'}%)",
|
||||
"value": f"{'%.2f' % (long_stop_dist*100)}%" if long_stop_dist else "N/A",
|
||||
},
|
||||
"support_alive_1h": {
|
||||
"pass": support_alive_1h,
|
||||
"desc": "1H 支撑最近被测试过且未跌破",
|
||||
"value": "有效" if support_alive_1h else "失效/未测试",
|
||||
},
|
||||
"strong_uptrend_4h": {
|
||||
"pass": strong_uptrend_4h,
|
||||
"desc": "4H 趋势强度（扩张中）",
|
||||
"value": f"{'%.2f' % (trend_strength_up_val*100)}%",
|
||||
},
|
||||
"trend_down_1d": {
|
||||
"pass": trend_down_1d,
|
||||
"desc": "D1 下降趋势（LH+LL 都在降低）",
|
||||
"value": "下降" if trend_down_1d else ("上升" if trend_up_1d else "震荡"),
|
||||
},
|
||||
"in_supply_1h": {
|
||||
"pass": in_supply_1h,
|
||||
"desc": "价格在 1H 供给区（zone 顶部 65%+）",
|
||||
"value": f"位置 {diagnosis['price_position_in_zone']}%" if diagnosis['price_position_in_zone'] is not None else "N/A",
|
||||
},
|
||||
"short_stop_distance": {
|
||||
"pass": short_dist_ok,
|
||||
"desc": f"入场距阻力距离 ({'%.2f' % (short_stop_dist*100) if short_stop_dist else 'N/A'}%)",
|
||||
"value": f"{'%.2f' % (short_stop_dist*100)}%" if short_stop_dist else "N/A",
|
||||
},
|
||||
"resistance_alive_1h": {
|
||||
"pass": resistance_alive_1h,
|
||||
"desc": "1H 阻力最近被测试过且未突破",
|
||||
"value": "有效" if resistance_alive_1h else "失效/未测试",
|
||||
},
|
||||
"strong_downtrend_4h": {
|
||||
"pass": strong_downtrend_4h,
|
||||
"desc": "4H 下降趋势强度（扩张中）",
|
||||
"value": f"{'%.2f' % (trend_strength_down_val*100)}%",
|
||||
},
|
||||
}
|
||||
|
||||
# 入场可行性结论（1H S/R + 4H 趋势）
|
||||
long_ok = all([
|
||||
trend_up_1d, in_demand_1h, long_dist_ok,
|
||||
support_alive_1h, strong_uptrend_4h,
|
||||
])
|
||||
short_ok = all([
|
||||
trend_down_1d, in_supply_1h, short_dist_ok,
|
||||
resistance_alive_1h, strong_downtrend_4h,
|
||||
])
|
||||
|
||||
return {
|
||||
"timestamp": pd.Timestamp.now().isoformat(),
|
||||
"current_price": round(current_price, 2),
|
||||
"diagnosis": diagnosis,
|
||||
"filters": filters,
|
||||
"can_enter_long": long_ok,
|
||||
"can_enter_short": short_ok,
|
||||
"trend_1d": "up" if trend_up_1d else ("down" if trend_down_1d else "neutral"),
|
||||
"candle_1h": {
|
||||
"bullish_pinbar": bullish_pinbar,
|
||||
"bearish_pinbar": bearish_pinbar,
|
||||
"bullish_signal": bullish_signal,
|
||||
"bearish_signal": bearish_signal,
|
||||
},
|
||||
"swing_points": {
|
||||
"highs": [{"time": t, "price": p} for t, p in zip(swing_highs_1h[-10:], swing_high_prices_1h[-10:])],
|
||||
"lows": [{"time": t, "price": p} for t, p in zip(swing_lows_1h[-10:], swing_low_prices_1h[-10:])],
|
||||
},
|
||||
"trend_strength_4h": {
|
||||
"up": round(trend_strength_up_val * 100, 2),
|
||||
"down": round(trend_strength_down_val * 100, 2),
|
||||
},
|
||||
}
|
||||
498
docs/backtest-pitfalls.md
Normal file
498
docs/backtest-pitfalls.md
Normal file
@ -0,0 +1,498 @@
|
||||
# Freqtrade 回测部署踩坑记录
|
||||
|
||||
> 记录于 2026-06-07 | PriceActionStrategy v0.2 首次回测
|
||||
> 环境：Docker freqtrade (d:\ft_userdata\)，ETH/USDT，Binance
|
||||
|
||||
---
|
||||
|
||||
## 一、配置文件 (config.json) 的隐性必需字段
|
||||
|
||||
Freqtrade 的配置校验非常严格，即使某些模块设为 `disabled`，也**必须提供完整的字段结构**。
|
||||
以下字段在首次部署时缺失，逐一触发报错：
|
||||
|
||||
### 1. telegram 模块
|
||||
|
||||
**错误**：只写了 `"enabled": false`，缺少 `token` 和 `chat_id`
|
||||
|
||||
```json
|
||||
// ❌ 错误（会报 schema 校验失败）
|
||||
"telegram": {
|
||||
"enabled": false
|
||||
}
|
||||
|
||||
// ✅ 正确（即使 disabled 也要提供空值）
|
||||
"telegram": {
|
||||
"enabled": false,
|
||||
"token": "",
|
||||
"chat_id": ""
|
||||
}
|
||||
```
|
||||
|
||||
### 2. api_server 模块
|
||||
|
||||
**错误**：只写了 `"enabled": false`
|
||||
|
||||
```json
|
||||
// ❌ 错误
|
||||
"api_server": {
|
||||
"enabled": false
|
||||
}
|
||||
|
||||
// ✅ 正确
|
||||
"api_server": {
|
||||
"enabled": false,
|
||||
"listen_ip_address": "0.0.0.0",
|
||||
"listen_port": 8080,
|
||||
"username": "freqtrader",
|
||||
"password": "password",
|
||||
"jwt_secret_key": "somethingRandom123"
|
||||
}
|
||||
```
|
||||
|
||||
### 3. entry_pricing / exit_pricing
|
||||
|
||||
回测配置文件**必须显式声明**这两个模块。
|
||||
|
||||
```json
|
||||
// ✅ 回测 config 必须包含
|
||||
"entry_pricing": {
|
||||
"price_side": "same",
|
||||
"use_order_book": false,
|
||||
"order_book_top": 1,
|
||||
"price_last_balance": 0.0,
|
||||
"check_depth_of_market": {
|
||||
"enabled": false,
|
||||
"bids_to_ask_delta": 1
|
||||
}
|
||||
},
|
||||
"exit_pricing": {
|
||||
"price_side": "same",
|
||||
"use_order_book": false,
|
||||
"order_book_top": 1
|
||||
}
|
||||
```
|
||||
|
||||
### 4. ccxt_config 中的 proxies
|
||||
|
||||
在 Docker 环境下，`proxies` 字段会导致配置解析异常。**不要**在 ccxt_config 中加入代理配置，除非明确知道 freqtrade 版本支持。
|
||||
|
||||
```json
|
||||
// ❌ 导致配置解析错误
|
||||
"ccxt_config": {
|
||||
"enableRateLimit": true,
|
||||
"proxies": { ... }
|
||||
}
|
||||
|
||||
// ✅ 正确的极简配置
|
||||
"ccxt_config": {
|
||||
"enableRateLimit": true
|
||||
},
|
||||
"ccxt_async_config": {
|
||||
"enableRateLimit": true
|
||||
}
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 二、策略文件的强制属性
|
||||
|
||||
Freqtrade 要求每个策略必须声明以下属性，否则回测直接失败：
|
||||
|
||||
```python
|
||||
class PriceActionStrategy(IStrategy):
|
||||
# ── 必需属性 ──
|
||||
timeframe = "5m" # 主时间框架
|
||||
can_short = True # 是否支持做空（spot 模式必须为 False）
|
||||
stoploss = -0.10 # 硬止损比例（即使有 custom_stoploss 也要声明）
|
||||
use_custom_stoploss = True # 是否使用动态止损
|
||||
minimal_roi = {"0": 100} # 时间止盈表（不用就写 {"0": 100}）
|
||||
max_open_trades = 1 # 最大同时持仓数
|
||||
|
||||
# ── 必需方法 ──
|
||||
def populate_indicators(self, dataframe, metadata): ...
|
||||
def populate_entry_trend(self, dataframe, metadata): ...
|
||||
def populate_exit_trend(self, dataframe, metadata): ...
|
||||
```
|
||||
|
||||
| 属性 | 说明 | 缺失后果 |
|
||||
|------|------|---------|
|
||||
| `stoploss` | 硬止损比例，如 `-0.10` = 10% | 回测直接报错退出 |
|
||||
| `minimal_roi` | 时间止盈表，不用则写 `{"0": 100}` | 回测直接报错退出 |
|
||||
| `use_custom_stoploss` | 有 `custom_stoploss()` 方法时必须设为 True | 方法不被调用 |
|
||||
|
||||
### Spot vs Futures 的 can_short 差异
|
||||
|
||||
| 模式 | `can_short` | 说明 |
|
||||
|------|------------|------|
|
||||
| `trading_mode: "spot"` | **必须是 False** | 现货不支持做空 |
|
||||
| `trading_mode: "futures"` | 可以 True | 合约支持双向 |
|
||||
|
||||
**最佳实践**：策略中保持 `can_short = True`（你的实盘方向），创建单独的回测配置指定 `trading_mode: "futures"` 来测完整逻辑。如果必须用现货数据跑 spot 回测，临时改为 False。
|
||||
|
||||
---
|
||||
|
||||
## 三、多时间框架合并的数据清洗陷阱
|
||||
|
||||
### 问题：merge_informative_pair 导致 NaN
|
||||
|
||||
当使用 `merge_informative_pair()` 合并 D1/1H 数据到 5M 主框架时，**前 N 根 K 线的衍生列（布尔值、EMA、Swing 点）全是 NaN**。
|
||||
|
||||
```python
|
||||
# merge_informative_pair 返回的 dataframe 前面有 NaN 行
|
||||
dataframe = merge_informative_pair(
|
||||
dataframe, daily, self.timeframe, "1d", ffill=True,
|
||||
)
|
||||
# 结果：前几十行中 trend_up_1d、bullish_pinbar_1h 等列为 NaN
|
||||
```
|
||||
|
||||
**错误表现**：`ValueError: The truth value of a Series is ambiguous` 或 NaN 通过布尔条件传递导致崩溃。
|
||||
|
||||
**修复方案**：在 `populate_indicators()` 返回前，显式填充所有布尔列：
|
||||
|
||||
```python
|
||||
def populate_indicators(self, dataframe, metadata):
|
||||
# ... 所有指标计算 ...
|
||||
|
||||
# ⚠️ 关键：填充合并后的 NaN 布尔列
|
||||
bool_cols = [
|
||||
"trend_up_1d", "trend_down_1d",
|
||||
"trend_up_1h", "trend_down_1h",
|
||||
"bullish_pinbar", "bearish_pinbar",
|
||||
"bullish_engulfing", "bearish_engulfing",
|
||||
"volume_surge",
|
||||
]
|
||||
for col in bool_cols:
|
||||
if col in dataframe.columns:
|
||||
dataframe[col] = dataframe[col].fillna(False).infer_objects(copy=False)
|
||||
|
||||
return dataframe
|
||||
```
|
||||
|
||||
### 列名冲突问题
|
||||
|
||||
`merge_informative_pair` 默认 `append_timeframe=True`（给合并的列加 `_1d`、`_1h` 后缀）。**不要**传 `append_timeframe=False`，会导致源列被覆盖。后续引用时记得带后缀：
|
||||
|
||||
```python
|
||||
# ✅ 正确引用
|
||||
dataframe["trend_up_1d"] # 来自日线的趋势判断
|
||||
dataframe["trend_up_1h"] # 来自1H的趋势判断
|
||||
dataframe["swing_low_1h"] # 来自1H的Swing Low
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 四、回测执行命令模板
|
||||
|
||||
```bash
|
||||
cd d:/ft_userdata
|
||||
docker compose run --rm freqtrade backtesting \
|
||||
--config user_data/config_backtest.json \
|
||||
--strategy PriceActionStrategy \
|
||||
--timerange 20250101-
|
||||
```
|
||||
|
||||
### 必备前置步骤
|
||||
|
||||
1. **策略文件已复制到 Docker 目录**：
|
||||
```bash
|
||||
cp user_data/strategies/price_action_strategy.py \
|
||||
d:/ft_userdata/user_data/strategies/PriceActionStrategy.py
|
||||
```
|
||||
2. **历史数据已下载**（`d:/ft_userdata/user_data/data/binance/` 下有对应的 JSON 文件）
|
||||
3. **回测配置文件 `config_backtest.json` 已就绪**，包含所有必需字段
|
||||
|
||||
---
|
||||
|
||||
## 五、完整的最小可行回测配置
|
||||
|
||||
```json
|
||||
{
|
||||
"max_open_trades": 1,
|
||||
"stake_currency": "USDT",
|
||||
"stake_amount": 10000,
|
||||
"tradable_balance_ratio": 0.99,
|
||||
"fiat_display_currency": "USD",
|
||||
"dry_run": true,
|
||||
"trading_mode": "spot",
|
||||
"margin_mode": "",
|
||||
"exchange": {
|
||||
"name": "binance",
|
||||
"key": "",
|
||||
"secret": "",
|
||||
"ccxt_config": {
|
||||
"enableRateLimit": true
|
||||
},
|
||||
"ccxt_async_config": {
|
||||
"enableRateLimit": true
|
||||
}
|
||||
},
|
||||
"pairlists": [
|
||||
{"method": "StaticPairList"}
|
||||
],
|
||||
"telegram": {
|
||||
"enabled": false,
|
||||
"token": "",
|
||||
"chat_id": ""
|
||||
},
|
||||
"api_server": {
|
||||
"enabled": false,
|
||||
"listen_ip_address": "0.0.0.0",
|
||||
"listen_port": 8080,
|
||||
"username": "freqtrader",
|
||||
"password": "password",
|
||||
"jwt_secret_key": "somethingRandom123"
|
||||
},
|
||||
"bot_name": "backtest",
|
||||
"entry_pricing": {
|
||||
"price_side": "same",
|
||||
"use_order_book": false,
|
||||
"order_book_top": 1,
|
||||
"price_last_balance": 0.0,
|
||||
"check_depth_of_market": {
|
||||
"enabled": false,
|
||||
"bids_to_ask_delta": 1
|
||||
}
|
||||
},
|
||||
"exit_pricing": {
|
||||
"price_side": "same",
|
||||
"use_order_book": false,
|
||||
"order_book_top": 1
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 六、错误顺序时间线
|
||||
|
||||
按实际触发顺序排列，方便快速定位：
|
||||
|
||||
| 序号 | 错误 | 根因 | 修复 |
|
||||
|------|------|------|------|
|
||||
| 1 | 配置解析错误 | `ccxt_config` 中含 `proxies` 字段 | 移除 proxies |
|
||||
| 2 | Schema 校验失败 | `telegram` 缺少 `token`/`chat_id` | 补全空值字段 |
|
||||
| 3 | Schema 校验失败 | `api_server` 缺少完整结构 | 补全所有必需字段 |
|
||||
| 4 | Schema 校验失败 | 缺少 `entry_pricing`/`exit_pricing` | 添加两个模块 |
|
||||
| 5 | 策略加载失败 | spot 模式不支持 `can_short=True` | 临时设 False |
|
||||
| 6 | 策略校验失败 | 缺少 `stoploss` 属性 | 添加 stoploss |
|
||||
| 7 | 策略校验失败 | 有 `custom_stoploss()` 但未声明 `use_custom_stoploss` | 设为 True |
|
||||
| 8 | 策略校验失败 | 缺少 `minimal_roi` | 添加 `{"0": 100}` |
|
||||
| 9 | NaN 导致崩溃 | `merge_informative_pair` 后布尔列为 NaN | fillna(False) 清洗 |
|
||||
| **✅** | **回测成功** | — | — |
|
||||
|
||||
---
|
||||
|
||||
## 七、Freqtrade 2026.2 Binance Futures 回测专用坑
|
||||
|
||||
### 问题：`Ticker pricing not available for Binance`
|
||||
|
||||
**触发条件**：`trading_mode: "futures"` + `use_order_book: false`
|
||||
|
||||
**根因**：`freqtrade/exchange/binance.py` 第 57 行：
|
||||
```python
|
||||
_ft_has_futures = {
|
||||
"tickers_have_price": False, # ← Binance futures 不支持 ticker.price
|
||||
}
|
||||
```
|
||||
`validate_pricing()` 检查 `self._ft_has["tickers_have_price"]`，若为 False 且 `use_order_book=false`，直接抛异常。
|
||||
|
||||
**修复方案**：futures 模式下**必须**设置：
|
||||
```json
|
||||
"entry_pricing": { "use_order_book": true },
|
||||
"exit_pricing": { "use_order_book": true }
|
||||
```
|
||||
|
||||
### 问题：`SampleStrategy` 也无法运行 futures 回测
|
||||
|
||||
**确认**：不是策略问题，是 freqtrade 2026.2 的 Binance futures 初始化 Bug。
|
||||
|
||||
**解决方案**：
|
||||
1. `use_order_book: true`（绕过 ticker 检查）
|
||||
2. 下载 mark price 数据：`download-data --trading-mode futures`
|
||||
3. pair 名称必须带结算货币后缀：`ETH/USDT:USDT`
|
||||
|
||||
### 最小可行 futures 回测配置
|
||||
|
||||
```json
|
||||
{
|
||||
"trading_mode": "futures",
|
||||
"margin_mode": "cross",
|
||||
"liquidation_buffer": 0.05,
|
||||
"exchange": {
|
||||
"name": "binance",
|
||||
"key": "",
|
||||
"secret": "",
|
||||
"ccxt_config": {
|
||||
"enableRateLimit": true,
|
||||
"options": {"defaultType": "swap"}
|
||||
},
|
||||
"pair_whitelist": ["ETH/USDT:USDT"],
|
||||
"pair_blacklist": []
|
||||
},
|
||||
"entry_pricing": {
|
||||
"use_order_book": true, // ← futures 必须 true
|
||||
"order_book_top": 1
|
||||
},
|
||||
"exit_pricing": {
|
||||
"use_order_book": true, // ← futures 必须 true
|
||||
"order_book_top": 1
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
### v1.1 回测结果（首次 futures 成功）
|
||||
|
||||
| 指标 | 数值 |
|
||||
|------|------|
|
||||
| 交易笔数 | 65 |
|
||||
| 总盈亏 | +61.52% |
|
||||
| 做多利润 | +5,193 USDT |
|
||||
| 做空利润 | +959 USDT |
|
||||
| 硬止损问题 | 31 笔全部亏损，-8,839 USDT |
|
||||
|
||||
**教训**：futures 可以做空，熊市不再被动挨打。但硬止损位置不对（放在结构失效点太远），v1.2 改为 Entry Candle 失效点。
|
||||
|
||||
| **✅** | **v1.1 futures 回测成功** | — | — |
|
||||
|
||||
---
|
||||
|
||||
## 八、v1.2 止损逻辑 Bug 分析（关键经验）
|
||||
|
||||
### v1.2 回测结果回顾
|
||||
|
||||
| 指标 | 数值 |
|
||||
|------|------|
|
||||
| 交易笔数 | 100 |
|
||||
| 硬止损出场 | **50 笔，全部亏损**，-8,468 USDT |
|
||||
| trailing_stop 出场 | 46 笔，34.8% 胜率，+13,671 USDT |
|
||||
|
||||
**50 笔硬止损全部亏损 = 止损逻辑根本性失效。**
|
||||
|
||||
### 根因 #1（致命）：`return None` 退回到 25% 宽止损
|
||||
|
||||
```python
|
||||
# v1.2 custom_stoploss 末尾（第 559 行）
|
||||
return None # ← 致命错误！
|
||||
```
|
||||
|
||||
`custom_stoploss` 返回 `None` 时，freqtrade **忽略自定义止损，使用类属性 `stoploss`**。
|
||||
v1.2 的 `stoploss = -0.25`（25% 硬止损），导致所有找不到 Entry Candle 的交易都用 25% 宽止损，"硬扛单"。
|
||||
|
||||
**修复**：`custom_stoploss` **永远不要返回 `None`**，始终返回显式止损比率。
|
||||
|
||||
### 根因 #2：Entry Candle 查找逻辑脆弱
|
||||
|
||||
```python
|
||||
# v1.2 的查找方式（第 507-519 行）
|
||||
entry_mask = (
|
||||
(dataframe["potential_entry_low"].notna())
|
||||
| (dataframe["potential_entry_high"].notna())
|
||||
)
|
||||
entry_candidates = dataframe[
|
||||
entry_mask
|
||||
& (dataframe["date"] <= trade.open_date_utc + timedelta(hours=1))
|
||||
& (dataframe["date"] >= trade.open_date_utc - timedelta(hours=1))
|
||||
]
|
||||
```
|
||||
|
||||
问题：
|
||||
1. `potential_entry_low` 标记了**所有**信号 K 线，不是触发这笔交易的那个
|
||||
2. ±1 小时窗口内有多个信号时，`iloc[-1]` 取的是最后一个，不一定正确
|
||||
3. 时区稍有偏差（纳秒级）就找不到，直接掉进 `return None`
|
||||
|
||||
**结论**：在 `custom_stoploss` 里查找 Entry Candle 是不可靠的，必须换方法。
|
||||
|
||||
### 根因 #3：阶段二 `pass` 后没有设止损
|
||||
|
||||
```python
|
||||
# v1.2 第 528-539 行
|
||||
if current_profit > self.profit_to_structure_sl_pct.value:
|
||||
pass # 继续到阶段二
|
||||
else:
|
||||
return max(sl_ratio, -0.25)
|
||||
# ← 如果阶段二条件不满足，函数一路走到 return None！
|
||||
```
|
||||
|
||||
盈利超阈值后跳到阶段二，但如果 `resistance_4h` / `support_4h` 为 NaN，阶段二没有任何 `return`，最终落到 `return None`。
|
||||
|
||||
### v1.3 修复方案
|
||||
|
||||
**弃用 Entry Candle 查找，改用 ATR 动态止损**：
|
||||
|
||||
```python
|
||||
# v1.3 custom_stoploss 核心逻辑
|
||||
atr = last_candle['atr_1h']
|
||||
|
||||
if not trade.is_short:
|
||||
if current_profit <= 0.01:
|
||||
sl_price = open_rate - atr * 1.0 # 阶段一：紧止损
|
||||
elif current_profit <= 0.02:
|
||||
sl_price = open_rate * 0.999 # 阶段二：保本
|
||||
else:
|
||||
sl_price = current_rate - atr * 1.0 # 阶段三：追踪止损
|
||||
sl_ratio = (sl_price / current_rate) - 1.0
|
||||
return max(sl_ratio, -0.05) # 永不返回 None！
|
||||
```
|
||||
|
||||
**为什么用 ATR 而不是 Entry Candle**：
|
||||
- ATR 来自当前 K 线数据，100% 可靠，不需要查找历史
|
||||
- ATR 是波动率的自适应指标，低波动时止损紧，高波动时止损宽
|
||||
- 符合"早早认输"哲学：价格反向跑 1 ATR 就止损，不硬扛
|
||||
|
||||
### 经验总结
|
||||
|
||||
| 规则 | 说明 |
|
||||
|------|------|
|
||||
| ❌ 永不 `return None` | 始终返回显式止损比率，宁可紧不要松 |
|
||||
| ✅ 用 ATR 做动态止损 | 可靠、自适应、不依赖历史 K 线查找 |
|
||||
| ✅ 三阶段设计 | 紧止损 → 保本 → 追踪，符合趋势交易逻辑 |
|
||||
| ✅ `stoploss` 属性设安全网 | `-0.05`（5%）防止极端行情，但 `custom_stoploss` 应永远先于此触发 |
|
||||
|
||||
| **✅** | **v1.3 止损逻辑重写完成** | — | — |
|
||||
|
||||
---
|
||||
|
||||
## 九、Docker 网络与代理环境（高频踩坑区）
|
||||
|
||||
### 核心事实
|
||||
|
||||
**freqtrade 2026.2 在启动回测时强制执行 `reload_markets`**，即使回测使用的全部是本地 OHLCV 数据。这意味着 **Docker 容器必须有外网访问能力**，否则任何操作都会在 `_load_async_markets` 阶段失败。
|
||||
|
||||
```
|
||||
freqtrade.exceptions.TemporaryError: Error in reload_markets due to ExchangeNotAvailable.
|
||||
Message: binance GET https://api.binance.com/api/v3/exchangeInfo
|
||||
```
|
||||
|
||||
### 已验证可工作的 docker-compose.yml 配置
|
||||
|
||||
```yaml
|
||||
services:
|
||||
freqtrade:
|
||||
environment:
|
||||
- HTTP_PROXY=http://127.0.0.1:7890
|
||||
- HTTPS_PROXY=http://127.0.0.1:7890
|
||||
- NO_PROXY=localhost,127.0.0.1
|
||||
- TZ=Asia/Shanghai
|
||||
```
|
||||
|
||||
**关键点**：
|
||||
- `HTTP_PROXY` 和 `HTTPS_PROXY` **必须同时设置**，缺一不可
|
||||
- `NO_PROXY` 中**不要包含 `api.binance.com`**——Binance 必须走代理
|
||||
- 端口号 `7890` 是用户 Clash 代理的默认端口
|
||||
|
||||
### 故障排查清单
|
||||
|
||||
| 现象 | 诊断方法 | 常见原因 |
|
||||
|------|---------|---------|
|
||||
| `ExchangeNotAvailable` | `docker run --rm curlimages/curl curl --proxy http://127.0.0.1:7890 https://api.binance.com/api/v3/ping` | 代理服务未运行 |
|
||||
| 容器连不上代理 | 从宿主机 `curl http://127.0.0.1:7890` | 代理端口不对或服务挂了 |
|
||||
| 配置正确但连不上 | 检查 `NO_PROXY` 是否误包含 `api.binance.com` | Binance 被排除在代理外 |
|
||||
|
||||
### 经验教训
|
||||
|
||||
1. **每次跑回测前**，确认代理服务（Clash/V2Ray）在运行
|
||||
2. **不要随意修改 docker-compose.yml 的代理配置**——它一旦配置正确就不该动
|
||||
3. **区分服务不可用 vs 配置错误**：同样配置之前能跑通现在不能 → 服务的锅不是配置的锅
|
||||
4. **在踩坑文档中固化已验证的 docker-compose.yml**，不要每次手动改
|
||||
|
||||
| **✅** | **网络/代理经验固化完成** | — | — |
|
||||
198
docs/stop_loss_analysis_report.md
Normal file
198
docs/stop_loss_analysis_report.md
Normal file
@ -0,0 +1,198 @@
|
||||
# v1.6 止损交易深度分析报告
|
||||
|
||||
> 分析时间：2026-06-08 | 数据：ETH/USDT 2022-01-01 ~ 2025-08-17 全周期回测
|
||||
> 总交易 152 笔，其中 stop_loss 退出 73 笔（全部亏损，胜率 0%）
|
||||
|
||||
---
|
||||
|
||||
## 一、核心发现
|
||||
|
||||
### 发现 1：**50.7% 的止损交易在入场时 H4 趋势方向就不对！**
|
||||
|
||||
这是最震撼的发现。73 笔止损交易中，有 37 笔（50.7%）在入场时 H4 级别趋势方向与交易方向不一致：
|
||||
|
||||
| 方向 | H4趋势不一致 | 典型场景 |
|
||||
|------|:----------:|---------|
|
||||
| LONG 止损 | 16/37 (43%) | D1看涨，但4H正在下跌——在下跌中的回调做多 |
|
||||
| SHORT 止损 | 21/36 (58%) | D1看跌，但4H正在上涨——在上涨中的反弹做空 |
|
||||
|
||||
**这意味着：** 策略的入场逻辑有结构性缺陷。当前入场条件只检查 D1 趋势方向，但**没有要求 H4 趋势与交易方向一致**。
|
||||
|
||||
### 案例说明
|
||||
|
||||
```
|
||||
#2: 2022-04-07 LONG
|
||||
D1趋势=UP ✓ → "大方向对"
|
||||
H4趋势=DOWN ✗ → "但4H正在跌"
|
||||
结果：支撑被击穿，-1.6%
|
||||
|
||||
#48: 2024-05-27 SHORT
|
||||
D1趋势=UP ✗ → "D1就不对！"
|
||||
H4趋势=UP ✗ → "4H也在涨"
|
||||
结果：3小时止损，-1.0%
|
||||
```
|
||||
|
||||
### 发现 2：**26 笔（35.6%）止损是因为支撑/阻力被直接击穿**
|
||||
|
||||
- LONG: 13 笔价格跌破支撑（PRICE_BROKE_SUPPORT）
|
||||
- SHORT: 13 笔价格突破阻力（PRICE_BROKE_RESISTANCE）
|
||||
|
||||
这些交易的特点是：入场时一切条件满足，但随后价格直接穿过 S/R 位。这是"真结构被否定"的情况——市场决定这个 S/R 不再有效。
|
||||
|
||||
### 发现 3：**15 笔（20.5%）止损是因为 S/R 位被新的 Swing Point 推着走**
|
||||
|
||||
- 支撑被新的 Swing Low 向下推移（LONG 做多 → 止损位下移 → 更大的亏损空间）
|
||||
- 阻力被新的 Swing High 向上推移（SHORT 做空 → 止损位上移 → 更大的亏损空间）
|
||||
|
||||
典型如 #13 (2022-09-19)：支撑位在交易期间**下移了 4.6%**，说明市场在持续创新低。
|
||||
|
||||
### 发现 4：**"活支撑/阻力"过滤器漏了 4 笔死支撑**
|
||||
|
||||
v1.6 的 `support_alive` 过滤器只捕捉到最近 3 根 4H bar 内的测试，但仍漏了 4 笔（5.5%）。这些支撑在入场时就没被验证过。
|
||||
|
||||
### 发现 5：**快止损 vs 慢流血 — 两种不同的失败模式**
|
||||
|
||||
| 类型 | 数量 | 平均亏损 | 特征 |
|
||||
|------|:----:|:-------:|------|
|
||||
| 快止损 (<6h) | 11笔 | -1.14% | 入场即错，价格立即反向 |
|
||||
| 慢流血 (>72h) | 19笔 | -2.21% | 价格先小幅有利，然后趋势反转 |
|
||||
|
||||
快止损大部分是 H4 趋势不一致导致（如 #48、#52）。慢流血则更复杂——有的是 S/R 位被慢慢侵蚀，有的是趋势反转。
|
||||
|
||||
---
|
||||
|
||||
## 二、分类统计
|
||||
|
||||
### 按根因分类
|
||||
|
||||
| 根因 | 数量 | 占比 | 可修复性 |
|
||||
|------|:----:|:----:|:-------:|
|
||||
| **H4趋势不一致** | 37 | 50.7% | ✅ 高 — 加入 H4 趋势过滤器 |
|
||||
| **价格直接击穿S/R** | 26 | 35.6% | ⚠️ 中 — 需要更强的S/R确认 |
|
||||
| **S/R位移** | 15 | 20.5% | ⚠️ 中 — 部分可提前识别 |
|
||||
| **不在正确区域** | 12 | 16.4% | ✅ 高 — 收紧区域判断 |
|
||||
| **死支撑/阻力** | 4 | 5.5% | ✅ 已过滤（v1.6已解决大部分） |
|
||||
| **不明原因** | 8 | 11.0% | 🔬 需要逐笔人工审查 |
|
||||
|
||||
> 注：一笔交易可能有多重根因，总和超过 100%。
|
||||
|
||||
### 按年度分布
|
||||
|
||||
| 年份 | 止损数 | 总亏损 | 平均亏损 | 市场环境 |
|
||||
|------|:-----:|:------:|:------:|:-------:|
|
||||
| 2022 | 21 | -$4,516 | -2.56% | 熊市 |
|
||||
| 2023 | 20 | -$5,030 | -1.58% | 恢复期 |
|
||||
| 2024 | 18 | -$21,772 | -2.16% | 牛市 |
|
||||
| 2025 | 14 | -$36,062 | -2.14% | 震荡市 |
|
||||
|
||||
> 2025 年止损亏损金额最大，因为本金滚大了（复利效应），但平均亏损%与其他年份相当。
|
||||
|
||||
---
|
||||
|
||||
## 三、可修复性评估
|
||||
|
||||
### ✅ 高优先级（大概率能修复）
|
||||
|
||||
**1. 加入 H4 趋势一致性过滤器**
|
||||
|
||||
这是最大的单一改进机会。当前逻辑：
|
||||
```python
|
||||
# 当前：只检查 D1 趋势
|
||||
long_base = trend_up_1d & in_demand_4h & bullish_signal
|
||||
|
||||
# 改进：同时要求 H4 趋势一致
|
||||
long_base = trend_up_1d & trend_up_4h & in_demand_4h & bullish_signal
|
||||
```
|
||||
|
||||
**预期效果：** 可能过滤掉 37 笔中的大部分（但需要评估是否会误杀盈利交易）。
|
||||
|
||||
**2. 收紧区域判断**
|
||||
|
||||
当前 `in_demand` = 价格在支撑-阻力区间的下 35%。12 笔止损交易不在正确区域就入场了。可以：
|
||||
- 收紧到 25% 或 20%
|
||||
- 或要求价格更接近支撑/阻力位
|
||||
|
||||
### ⚠️ 中优先级（部分可修复）
|
||||
|
||||
**3. 强化 S/R 确认**
|
||||
|
||||
26 笔"价格击穿 S/R"中，有些可能是支撑/阻力本身质量不高。可以：
|
||||
- 要求 Swing Point 被测试 ≥2 次才有效
|
||||
- 要求 S/R 位形成时间 > N 根 bar（避免新形成的、未被验证的）
|
||||
- 考虑 S/R 位的"级别"——日线 Swing Point 比 4H Swing Point 更重要
|
||||
|
||||
**4. S/R 位移预警**
|
||||
|
||||
15 笔 S/R 位移中，如果在交易期间检测到"S/R 位正在移动"（新 Swing Point 形成），可以提前退出而非等到止损触发。
|
||||
|
||||
### 🔬 需要进一步研究
|
||||
|
||||
**5. "不明原因"的 8 笔**
|
||||
|
||||
这 8 笔交易（#5, #16, #19, #29, #36, #44, #66, #69）看起来一切条件都对——H4 趋势对、D1 趋势对、在正确区域、支撑有效——但还是止损了。需要逐笔查看当时的 K 线图来理解。这可能是策略的"固有成本"。
|
||||
|
||||
---
|
||||
|
||||
## 四、UNFAIR 问题（结构性缺陷）
|
||||
|
||||
### 问题 1：D1 和 H4 的脱节
|
||||
|
||||
当前策略的核心假设是"D1 定方向，4H 找位置"，但实际执行中：
|
||||
- D1 趋势变化很慢（几天到几周）
|
||||
- H4 趋势变化很快（几小时到几天）
|
||||
- 在 D1 上涨但 H4 下跌的窗口期（回调），策略会做多 → 然后被 H4 下跌趋势吞噬
|
||||
|
||||
**本质：** D1 趋势对 1H 入场来说太"远"了，中间的 4H 趋势才是 1H 入场最近的上层结构。
|
||||
|
||||
### 问题 2：S/R 识别只看最近 4 个 Swing Point
|
||||
|
||||
`_build_structure` 只保留最近 4 个 Swing High/Low 来判断趋势。这可能导致：
|
||||
- 在更大时间框架的结构位（比如日线级别的关键支撑）被忽略
|
||||
- 趋势判断过于短视
|
||||
|
||||
### 问题 3：custom_stoploss 的动态更新是一把双刃剑
|
||||
|
||||
当新的 Swing Low 形成时，`support_4h` 会更新为新低点，止损位下移 → 持仓承受更大的浮亏。这就是 S/R 位移的根源。
|
||||
|
||||
---
|
||||
|
||||
## 五、改进路线图
|
||||
|
||||
### 阶段 1：H4 趋势过滤器（预计影响最大）
|
||||
|
||||
```
|
||||
v1.6.3: 加入 H4 趋势一致性要求
|
||||
- LONG: D1趋势=UP AND H4趋势=UP
|
||||
- SHORT: D1趋势=DOWN AND H4趋势=DOWN
|
||||
- 回测验证：是否会误杀太多盈利交易？
|
||||
```
|
||||
|
||||
### 阶段 2：S/R 质量提升
|
||||
|
||||
```
|
||||
v1.6.4: 增强 S/R 确认
|
||||
- 要求 Swing Point 至少被回测 1 次
|
||||
- 要求 S/R 形成时间 ≥ 2 根 4H bar
|
||||
- 优先使用 D1 级别的 Swing Point（如果存在）
|
||||
```
|
||||
|
||||
### 阶段 3：动态退出优化
|
||||
|
||||
```
|
||||
v1.6.5: S/R 位移时的提前退出
|
||||
- 当新 Swing Point 形成且方向不利时 → 提前 exit
|
||||
- 而非等待价格触及已移动的止损位
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 六、数据文件
|
||||
|
||||
| 文件 | 说明 |
|
||||
|------|------|
|
||||
| `bt_v16_stop_losses.csv` | 73 笔止损交易的基础数据 |
|
||||
| `bt_v16_stop_loss_analysis.csv` | 含结构快照和分类的完整分析数据 |
|
||||
|
||||
---
|
||||
|
||||
*这份报告是接下来所有止损优化的基础。每个改进方向都可以独立验证、独立回测。*
|
||||
735
docs/strategy-v2.1-whitepaper.md
Normal file
735
docs/strategy-v2.1-whitepaper.md
Normal file
@ -0,0 +1,735 @@
|
||||
# Structure Flow Strategy v2.1 — 策略白皮书
|
||||
|
||||
> **版本**: v2.1 | **日期**: 2026-06-08 | **状态**: 生产 dry-run
|
||||
>
|
||||
> 本文档面向策略使用者本人，用于深入理解策略逻辑、回测表现和设计思路。
|
||||
|
||||
---
|
||||
|
||||
## 目录
|
||||
|
||||
1. [策略概览](#1-策略概览)
|
||||
2. [理论基础](#2-理论基础)
|
||||
3. [时间框架架构](#3-时间框架架构)
|
||||
4. [核心技术：Swing Point 检测](#4-核心技术swing-point-检测)
|
||||
5. [结构分析：趋势与支撑阻力](#5-结构分析趋势与支撑阻力)
|
||||
6. [K线形态：入场触发器](#6-k线形态入场触发器)
|
||||
7. [入场逻辑详解](#7-入场逻辑详解)
|
||||
8. [止损逻辑](#8-止损逻辑)
|
||||
9. [出场逻辑](#9-出场逻辑)
|
||||
10. [v2.1 核心创新：趋势强度过滤](#10-v21-核心创新趋势强度过滤)
|
||||
11. [参数说明](#11-参数说明)
|
||||
12. [回测表现](#12-回测表现)
|
||||
13. [演进历史](#13-演进历史)
|
||||
14. [已知局限](#14-已知局限)
|
||||
15. [部署信息](#15-部署信息)
|
||||
|
||||
---
|
||||
|
||||
## 1. 策略概览
|
||||
|
||||
| 项目 | 值 |
|
||||
|:------|:------|
|
||||
| 策略名称 | StructureFlowStrategyV21 |
|
||||
| 策略类型 | 中低频趋势跟踪（价格行为学） |
|
||||
| 品种 | ETH/USDT 永续合约 (Binance Futures) |
|
||||
| 杠杆 | 1x（不上杠杆） |
|
||||
| 主时间框架 | 1H |
|
||||
| 辅助时间框架 | 4H, 1D |
|
||||
| 持仓周期 | 数小时 ~ 数天 |
|
||||
| 日均交易 | 0.1 ~ 0.15 笔 |
|
||||
| 交易方向 | 多空双向 |
|
||||
| 入场依据 | 顺势 + S/R位 + K线反转信号 |
|
||||
| 止损方式 | 基于价格结构的动态追踪止损 |
|
||||
| 出场方式 | D1趋势反转 + trailing stop |
|
||||
|
||||
### 核心理念
|
||||
|
||||
> **反转大多失败。在趋势中，于支撑/阻力位顺势入场，用追踪止损让利润奔跑。**
|
||||
|
||||
策略不预测市场方向，只做一件事：**在明确的趋势中，等待价格回到关键 S/R 位，确认反转信号后入场，然后用结构止损保护仓位，让趋势把利润带走。**
|
||||
|
||||
---
|
||||
|
||||
## 2. 理论基础
|
||||
|
||||
### 2.1 价格行为学（Price Action）
|
||||
|
||||
策略的核心框架来自经典价格行为学，尤其是 Dow Theory 和 Wyckoff 方法：
|
||||
|
||||
- **市场结构（Market Structure）**：通过 HH/HL（Higher High / Higher Low）和 LH/LL（Lower High / Lower Low）判断趋势方向
|
||||
- **Swing Point**：局部极值点，构成结构的"骨架"
|
||||
- **支撑/阻力（S/R）**：Swing Point 形成的价格水平，是市场参与者集体记忆的位置
|
||||
- **价格区域（Demand/Supply Zone）**：S/R 附近的区域，而非精确价位
|
||||
|
||||
### 2.2 为什么不用指标
|
||||
|
||||
策略刻意**不使用任何技术指标**（EMA、MACD、RSI、ATR 等），原因：
|
||||
|
||||
1. **指标是价格的衍生品**——滞后于价格本身
|
||||
2. **指标参数需要调优**——容易过拟合
|
||||
3. **价格结构是本源信息**——HH/HL/LH/LL 直接告诉你市场在做什么
|
||||
4. **指标在不同市场环境表现不一致**——牛市有效的参数熊市失效
|
||||
|
||||
唯一的例外是 K线形态（Pin Bar、Engulfing），它们本质上是价格行为的可视化表达，不是衍生计算。
|
||||
|
||||
### 2.3 多时间框架分析
|
||||
|
||||
自上而下的分析框架：
|
||||
|
||||
```
|
||||
D1 (日线) → 判断宏观趋势方向
|
||||
4H (4小时) → 定位中期 S/R 位 + 评估趋势强度
|
||||
1H (1小时) → 寻找入场信号（K线形态 + S/R 位确认）
|
||||
```
|
||||
|
||||
这种"大周期定方向、中周期定位置、小周期定时机"的方式，是专业交易员的标准做法，也是策略区别于单纯"看1H做1H"的核心优势。
|
||||
|
||||
---
|
||||
|
||||
## 3. 时间框架架构
|
||||
|
||||
策略使用 freqtrade 的 `@informative` 装饰器加载三个时间框架：
|
||||
|
||||
```
|
||||
┌─────────────────────────────────────────┐
|
||||
│ 1D (日线) │
|
||||
│ populate_indicators_1d() │
|
||||
│ → Swing Point 检测 │
|
||||
│ → 结构分析（trend_up / trend_down） │
|
||||
│ → 输出到 1H: trend_up_1d, trend_down_1d │
|
||||
└──────────────┬──────────────────────────┘
|
||||
│ 宏观趋势方向
|
||||
┌──────────────▼──────────────────────────┐
|
||||
│ 4H (4小时) │
|
||||
│ populate_indicators_4h() │
|
||||
│ → Swing Point 检测 │
|
||||
│ → 结构分析（trend + S/R + zone） │
|
||||
│ → 活支撑/阻力检查 │
|
||||
│ → ★ v2.1 趋势强度评估 │
|
||||
│ → 输出到 1H: *_4h 系列字段 │
|
||||
└──────────────┬──────────────────────────┘
|
||||
│ S/R位置 + 趋势强度
|
||||
┌──────────────▼──────────────────────────┐
|
||||
│ 1H (1小时) │
|
||||
│ populate_indicators() │
|
||||
│ → K线形态检测（Pin Bar, Engulfing） │
|
||||
│ → 合并所有时间框架数据 │
|
||||
│ populate_entry_trend() │
|
||||
│ → 综合判断入场信号 │
|
||||
│ custom_stoploss() │
|
||||
│ → 动态结构止损 │
|
||||
└─────────────────────────────────────────┘
|
||||
```
|
||||
|
||||
三个时间框架的数据通过 freqtrade 自动 merge 到 1H 主时间框架，列名会自动加后缀（如 `trend_up_1d`, `support_4h`）。
|
||||
|
||||
---
|
||||
|
||||
## 4. 核心技术：Swing Point 检测
|
||||
|
||||
### 4.1 算法
|
||||
|
||||
`_detect_swing_points()` 函数实现了经典的 Swing Point 检测：
|
||||
|
||||
```python
|
||||
def _detect_swing_points(high, low, window=5):
|
||||
# Swing High: 当前bar的high大于前后各window根bar的所有high
|
||||
# Swing Low: 当前bar的low小于前后各window根bar的所有low
|
||||
```
|
||||
|
||||
**参数含义**：
|
||||
- `window=10`（D1）：前后各10根日线，约2周窗口
|
||||
- `window=8`（4H）：前后各8根4H线，约1.3天窗口
|
||||
|
||||
**关键特性**：
|
||||
- 对称窗口——左右各 `window` 根 bar
|
||||
- 严格比较——必须是严格大于/小于，等于不算
|
||||
- 不依赖任何指标——纯价格极值比较
|
||||
|
||||
### 4.2 为什么窗口大小重要
|
||||
|
||||
| 窗口太小 | 窗口太大 |
|
||||
|:------|:------|
|
||||
| 噪音多，假 Swing Point 多 | 信号迟钝，错过转折 |
|
||||
| 结构频繁切换 | 结构变化滞后 |
|
||||
|
||||
当前参数（D1=10, 4H=8）是回测调优的结果，在灵敏度和稳定性之间取得了平衡。
|
||||
|
||||
### 4.3 可视化理解
|
||||
|
||||
```
|
||||
价格
|
||||
↑
|
||||
│ SH₁ SH₂
|
||||
│ ╱ ╲ ╱ ╲
|
||||
│ ╱ ╲ ╱ ╲
|
||||
│ ╱ ╲ SH₃ ╱ ╲
|
||||
│ ╱ ╲ ╱ ╲ ╱ ╲
|
||||
│╱ ╲╱ ╲ ╲___
|
||||
│ SL₁ SL₂ SL₃
|
||||
└─────────────────────────────────→ 时间
|
||||
|
||||
SH = Swing High（局部高点）
|
||||
SL = Swing Low（局部低点）
|
||||
|
||||
趋势判断：
|
||||
SH₂ > SH₁ 且 SL₂ > SL₁ → 上升趋势（HH + HL）
|
||||
SH₃ < SH₂ 且 SL₃ < SL₂ → 下降趋势（LH + LL）
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 5. 结构分析：趋势与支撑阻力
|
||||
|
||||
### 5.1 `_build_structure()` 函数
|
||||
|
||||
这个函数是策略的"大脑"，接收 Swing Point 数据，输出以下信息：
|
||||
|
||||
| 输出字段 | 含义 |
|
||||
|:------|:------|
|
||||
| `trend_up` | 当前是否处于上升趋势（HH + HL） |
|
||||
| `trend_down` | 当前是否处于下降趋势（LH + LL） |
|
||||
| `support` | 最近的 Swing Low 价格（支撑位） |
|
||||
| `resistance` | 最近的 Swing High 价格（阻力位） |
|
||||
| `in_demand` | 价格是否在需求区（支撑位附近下方35%区域） |
|
||||
| `in_supply` | 价格是否在供给区（阻力位附近上方35%区域） |
|
||||
|
||||
### 5.2 趋势判断逻辑
|
||||
|
||||
```
|
||||
if 最近的SH > 前一个SH AND 最近的SL > 前一个SL:
|
||||
→ 上升趋势（HH + HL）
|
||||
elif 最近的SH < 前一个SH AND 最近的SL < 前一个SL:
|
||||
→ 下降趋势（LH + LL）
|
||||
else:
|
||||
→ 维持上一个趋势判断（避免噪音切换）
|
||||
```
|
||||
|
||||
**关键设计**：当 HH/HL 和 LH/LL 都不满足时（震荡），策略**保持上一个趋势方向**而非标记为"无趋势"。这避免了在震荡市中频繁切换方向。
|
||||
|
||||
### 5.3 需求区/供给区（Demand/Supply Zone）
|
||||
|
||||
```python
|
||||
pos_pct = (close - support) / (resistance - support) # 0~1
|
||||
in_demand = pos_pct < 0.35 # 价格在 S/R 区间下方 35%
|
||||
in_supply = pos_pct > 0.65 # 价格在 S/R 区间上方 65%
|
||||
```
|
||||
|
||||
这是一个相对位置判断——不是固定金额，而是相对于 S/R 区间宽度的比例。这意味着：
|
||||
- 窄区间：离 S/R 很近才算"靠近"
|
||||
- 宽区间：离 S/R 稍远也算"靠近"
|
||||
|
||||
### 5.4 活支撑/活阻力（v1.6 新增）
|
||||
|
||||
```python
|
||||
# 活支撑：价格触及支撑位（±0.5%），但收盘价仍在支撑上方
|
||||
touched_support = (low <= support * 1.005) & (low >= support * 0.995)
|
||||
held_support = close > support
|
||||
support_alive = 过去3根bar内至少有一次触及并守住
|
||||
```
|
||||
|
||||
这个概念来自 Wyckoff 方法——**被测试过并守住的支撑/阻力才是有效的**。未被测试的 S/R 位只是"理论上的"，被测试并守住的才是"实战验证的"。
|
||||
|
||||
---
|
||||
|
||||
## 6. K线形态：入场触发器
|
||||
|
||||
策略识别两种经典的价格行为反转形态：
|
||||
|
||||
### 6.1 Pin Bar（锤子线/射击之星）
|
||||
|
||||
```python
|
||||
pin_bar_wick_ratio = 0.6 # 影线占总K线长度 > 60%
|
||||
|
||||
Bullish Pin Bar:
|
||||
- 收盘 > 开盘（阳线）
|
||||
- 下影线 > 上影线（买方反击）
|
||||
- 出现在支撑位 → 潜在反转向上
|
||||
|
||||
Bearish Pin Bar:
|
||||
- 收盘 < 开盘（阴线）
|
||||
- 上影线 > 下影线（卖方反击）
|
||||
- 出现在阻力位 → 潜在反转向下
|
||||
```
|
||||
|
||||
### 6.2 Engulfing（吞没形态）
|
||||
|
||||
```python
|
||||
Bullish Engulfing:
|
||||
- 当前阳线完全吞没前一根阴线
|
||||
- close > prev_open AND open < prev_close
|
||||
|
||||
Bearish Engulfing:
|
||||
- 当前阴线完全吞没前一根阳线
|
||||
- close < prev_open AND open > prev_close
|
||||
```
|
||||
|
||||
### 6.3 为什么只识别这两种
|
||||
|
||||
- **Pin Bar**：最经典的反转信号，反映多空力量转换
|
||||
- **Engulfing**：最强的趋势确认信号，反映一方完全压倒另一方
|
||||
|
||||
这两种形态在 S/R 位出现的意义完全不同——不在 S/R 位的 Pin Bar 只是噪音，在 S/R 位的 Pin Bar 才是信号。策略通过结合 S/R 位来过滤噪音。
|
||||
|
||||
---
|
||||
|
||||
## 7. 入场逻辑详解
|
||||
|
||||
### 7.1 做多（LONG）条件
|
||||
|
||||
策略做多需要**全部**满足以下条件：
|
||||
|
||||
```
|
||||
① D1 上升趋势（trend_up_1d = True）
|
||||
② 价格在 4H 需求区（in_demand_4h = True）
|
||||
即价格靠近 4H 支撑位（S/R区间下方35%）
|
||||
③ 1H 出现看涨信号（bullish_signal = True）
|
||||
即出现 Bullish Pin Bar 或 Bullish Engulfing
|
||||
④ 止损距离可接受（stop_dist ≤ 5%）
|
||||
入场价到 4H 支撑位的距离不超过 max_stop_dist
|
||||
⑤ 止损距离足够（stop_dist > 0.3%）
|
||||
避免止损设太近被噪音震出
|
||||
⑥ 4H 支撑是活的（support_alive_4h = True）
|
||||
支撑位在过去3根bar内被测试并守住
|
||||
⑦ ★ v2.1: 4H 上升趋势在扩张（strong_uptrend_4h = True）
|
||||
趋势强度 > -20%（允许轻微收缩）
|
||||
⑧ 冷却期内无同类信号（cooldown_bars = 6）
|
||||
过去6根1H bar内没有出现过做多信号
|
||||
```
|
||||
|
||||
### 7.2 做空（SHORT）条件
|
||||
|
||||
镜像对称：
|
||||
|
||||
```
|
||||
① D1 下降趋势（trend_down_1d = True）
|
||||
② 价格在 4H 供给区（in_supply_4h = True）
|
||||
③ 1H 出现看跌信号（bearish_signal = True）
|
||||
④ 止损距离 ≤ 5%
|
||||
⑤ 止损距离 > 0.3%
|
||||
⑥ 4H 阻力是活的（resistance_alive_4h = True）
|
||||
⑦ ★ v2.1: 4H 下降趋势在扩张（strong_downtrend_4h = True）
|
||||
⑧ 冷却期 6 根 bar
|
||||
```
|
||||
|
||||
### 7.3 入场逻辑图
|
||||
|
||||
```
|
||||
┌──────────────┐
|
||||
│ D1 趋势确认 │
|
||||
│ 上升/下降？ │
|
||||
└──────┬───────┘
|
||||
│ ✅
|
||||
┌──────▼───────┐
|
||||
│ 靠近4H S/R？ │
|
||||
│ in_demand/ │
|
||||
│ in_supply？ │
|
||||
└──────┬───────┘
|
||||
│ ✅
|
||||
┌──────▼───────┐
|
||||
│ 1H 反转信号？│
|
||||
│ Pin Bar/ │
|
||||
│ Engulfing？ │
|
||||
└──────┬───────┘
|
||||
│ ✅
|
||||
┌──────▼───────┐
|
||||
│ 止损距离 │
|
||||
│ 0.3%~5%？ │
|
||||
└──────┬───────┘
|
||||
│ ✅
|
||||
┌──────▼───────┐
|
||||
│ S/R是活的？ │
|
||||
│ 被测试+守住？ │
|
||||
└──────┬───────┘
|
||||
│ ✅
|
||||
┌──────▼───────┐
|
||||
│ ★ 趋势在扩张？│ ← v2.1 新增
|
||||
│ strong_ │
|
||||
│ uptrend/ │
|
||||
│ downtrend？ │
|
||||
└──────┬───────┘
|
||||
│ ✅
|
||||
┌──────▼───────┐
|
||||
│ 冷却期检查 │
|
||||
│ 6根bar内无 │
|
||||
│ 同类信号？ │
|
||||
└──────┬───────┘
|
||||
│ ✅
|
||||
┌──────▼───────┐
|
||||
│ ⚡ 入场！ │
|
||||
└──────────────┘
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 8. 止损逻辑
|
||||
|
||||
### 8.1 设计原则
|
||||
|
||||
策略使用 **custom_stoploss** 实现动态止损，基于价格结构而非固定百分比。
|
||||
|
||||
**核心理念**：止损位置由市场结构决定，不由固定金额或百分比决定。
|
||||
|
||||
### 8.2 止损算法
|
||||
|
||||
```python
|
||||
# 做多止损
|
||||
support = 4H最近的Swing Low（支撑位）
|
||||
sl_price = support × 0.999 # 支撑位下方 0.1%
|
||||
sl_ratio = (sl_price / current_rate) - 1.0
|
||||
return max(sl_ratio, -0.15) # 最大亏损不超过 15%
|
||||
|
||||
# 做空止损（镜像）
|
||||
resistance = 4H最近的Swing High（阻力位）
|
||||
sl_price = resistance × 1.001 # 阻力位上方 0.1%
|
||||
sl_ratio = 1.0 - (sl_price / current_rate)
|
||||
return min(sl_ratio, 0.15) # 最大亏损不超过 15%
|
||||
```
|
||||
|
||||
### 8.3 为什么用 0.1% 缓冲
|
||||
|
||||
经过 v1.7/v1.8 的多次尝试（2%、5%缓冲），0.1% 是最优的：
|
||||
|
||||
- **0.1%**：止损次数多，但单笔亏损小。关键是被止损后不影响后续交易，且 `trailing_stop_loss` 有更多机会发挥作用
|
||||
- **2~5%**：止损次数少，但单笔亏损大。`trailing_stop_loss` 的追踪效果被破坏
|
||||
|
||||
**核心认知**：这个策略的利润来源不是"少止损"，而是"让盈利交易跑起来"。大缓冲破坏了 trailing stop 的追踪效果。
|
||||
|
||||
### 8.4 -15% 硬止损
|
||||
|
||||
`stoploss = -0.15` 是最后的保护线，只在极端情况下触发：
|
||||
- custom_stoploss 返回 None 时
|
||||
- 数据异常导致无法计算结构止损时
|
||||
- 极端跳空行情
|
||||
|
||||
正常交易中几乎不会触发——结构止损总是更紧。
|
||||
|
||||
---
|
||||
|
||||
## 9. 出场逻辑
|
||||
|
||||
### 9.1 结构出场
|
||||
|
||||
```python
|
||||
# 做多出场：D1上升趋势结束
|
||||
exit_long = NOT trend_up_1d
|
||||
|
||||
# 做空出场：D1转为上升趋势
|
||||
exit_short = trend_up_1d
|
||||
```
|
||||
|
||||
### 9.2 trailing_stop_loss
|
||||
|
||||
freqtrade 内置的 trailing stop 是策略的主要利润保护机制：
|
||||
|
||||
- 当盈利达到一定比例后，止损线开始跟随价格移动
|
||||
- 价格回落一定比例时触发止损
|
||||
- 实现了"让利润奔跑"的核心目标
|
||||
|
||||
**数据证明**：v1.6 中 69.3% 的盈利交易通过 trailing_stop_loss 出场，这是策略利润的主要来源。
|
||||
|
||||
### 9.3 minimal_roi
|
||||
|
||||
```python
|
||||
minimal_roi = {"0": 100}
|
||||
```
|
||||
|
||||
策略不使用固定时间/收益比的止盈。所有利润保护交给 trailing stop。这意味着：
|
||||
- 没有"赚够了就平仓"的思维
|
||||
- 让趋势把利润带到它能带到的最远
|
||||
|
||||
---
|
||||
|
||||
## 10. v2.1 核心创新：趋势强度过滤
|
||||
|
||||
### 10.1 问题背景
|
||||
|
||||
v1.6 的痛点：在震荡市中，D1 趋势可能仍然是"上升"或"下降"（因为趋势判断只比较两个 Swing Point），但趋势本身的力度在减弱。这种"名义上的趋势"会导致在 S/R 位入场后被震荡出局。
|
||||
|
||||
### 10.2 趋势强度算法
|
||||
|
||||
v2.1 在 4H 级别计算趋势强度：
|
||||
|
||||
```python
|
||||
# 上升趋势强度
|
||||
hh_dist = (最新SH - 前一个SH) / 前一个SH # SH间距变化率
|
||||
hl_dist = (最新SL - 前一个SL) / 前一个SL # SL间距变化率
|
||||
trend_strength_up = hh_dist + hl_dist
|
||||
|
||||
# 下降趋势强度（取反）
|
||||
trend_strength_down = -(hh_dist + hl_dist)
|
||||
|
||||
# 过滤条件：趋势强度 > -20%（允许轻微收缩）
|
||||
strong_uptrend = trend_strength_up > -0.20
|
||||
strong_downtrend = trend_strength_down > -0.20
|
||||
```
|
||||
|
||||
### 10.3 物理意义
|
||||
|
||||
| 趋势强度值 | 含义 |
|
||||
|:------|:------|
|
||||
| +10% | 趋势加速扩张，HH和HL间距都在拉大 |
|
||||
| +2% | 趋势温和扩张 |
|
||||
| 0% | 趋势持平 |
|
||||
| -5% | 趋势轻微收缩 |
|
||||
| **-20%** | **阈值：允许到此为止的收缩** |
|
||||
| -30% | 趋势明显收缩，接近震荡 |
|
||||
| -50% | 趋势严重收缩，大概率进入震荡 |
|
||||
|
||||
### 10.4 为什么阈值是 -20%
|
||||
|
||||
回测参数调优结果。0%（只允许扩张）会过滤掉太多信号；-20% 在"过滤震荡市信号"和"保留趋势市信号"之间取得了最佳平衡。
|
||||
|
||||
### 10.5 v2.1 vs v1.6 效果对比
|
||||
|
||||
| 指标 | v1.6 | v2.1 | 变化 |
|
||||
|:------|:----:|:----:|:----:|
|
||||
| ETH 总收益 | +3659% | +4366% | +19.3% |
|
||||
| BTC 总收益 | +507% | +575% | +13.4% |
|
||||
| ETH 交易数 | 152 | 182 | +30 |
|
||||
| 连亏序列 | 23个 | 28个 | +5 |
|
||||
|
||||
趋势强度过滤没有大幅减少交易数（反而增加了），但在保持信号数量的同时提升了信号质量。这表明过滤器主要排除了"看起来可以但实际上不该入场"的信号，而不是简单粗暴地减少交易。
|
||||
|
||||
---
|
||||
|
||||
## 11. 参数说明
|
||||
|
||||
### 11.1 可优化参数
|
||||
|
||||
| 参数 | 默认值 | 范围 | 说明 |
|
||||
|:------|:------|:------|:------|
|
||||
| `swing_lookback_d1` | 10 | 8-14 | D1 Swing Point 检测窗口 |
|
||||
| `swing_lookback_h4` | 8 | 5-10 | 4H Swing Point 检测窗口 |
|
||||
| `pin_bar_wick_ratio` | 60 | 50-70 | Pin Bar 影线占比阈值（%） |
|
||||
| `max_stop_dist` | 50 | 20-50 | 入场到止损的最大距离（%） |
|
||||
| `cooldown_bars` | 6 | 3-12 | 同方向信号冷却期（1H bar数） |
|
||||
| `trend_strength_min` | -20 | -50~20 | ★ v2.1: 趋势强度最小阈值（%） |
|
||||
|
||||
### 11.2 固定参数
|
||||
|
||||
| 参数 | 值 | 说明 |
|
||||
|:------|:------|:------|
|
||||
| `stoploss` | -0.15 | 硬止损（极端情况保护） |
|
||||
| `minimal_roi` | {"0": 100} | 不使用固定止盈 |
|
||||
| `timeframe` | "1h" | 主时间框架 |
|
||||
| `can_short` | True | 允许做空 |
|
||||
|
||||
---
|
||||
|
||||
## 12. 回测表现
|
||||
|
||||
### 12.1 ETH/USDT 全周期（2022-01 ~ 2026-06）
|
||||
|
||||
| 指标 | v1.6 | v2.1 |
|
||||
|:------|:----:|:----:|
|
||||
| 总收益率 | +3659% | **+4366%** |
|
||||
| 总交易数 | 152 | 182 |
|
||||
| 胜率 | ~35% | ~36% |
|
||||
| Profit Factor | 3.56 | ~3.8 |
|
||||
| 最大连续亏损 | 8 笔 | 8 笔 |
|
||||
| 最长连亏天数 | 35 天 | ~30 天 |
|
||||
|
||||
### 12.2 BTC/USDT 全周期
|
||||
|
||||
| 指标 | v1.6 | v2.1 |
|
||||
|:------|:----:|:----:|
|
||||
| 总收益率 | +507% | **+575%** |
|
||||
| 总交易数 | 190 | ~200+ |
|
||||
|
||||
### 12.3 年度表现（ETH）
|
||||
|
||||
| 年份 | 市场环境 | 收益 | 说明 |
|
||||
|:------|:------|:------|:------|
|
||||
| 2022 | 熊市 | -11.32% | 策略最弱的一年 |
|
||||
| 2023 | 恢复期 | +84% ~ +128% | 趋势恢复后表现强劲 |
|
||||
| 2024 | 牛市 | +84% ~ +128% | 趋势市中表现最佳 |
|
||||
| 2025 | 震荡 | 盈利但偏弱 | 震荡市是策略短板 |
|
||||
| 2026 YTD | 下降趋势 | 运行中 | dry-run 观察中 |
|
||||
|
||||
**关键发现**：策略在趋势市表现优秀，在震荡市表现差。这符合价格行为策略的特征——没有趋势就没有利润。
|
||||
|
||||
### 12.4 回测配置
|
||||
|
||||
```json
|
||||
{
|
||||
"max_open_trades": 1,
|
||||
"dry_run_wallet": 10000,
|
||||
"stake_amount": "unlimited",
|
||||
"trading_mode": "futures",
|
||||
"margin_mode": "isolated"
|
||||
}
|
||||
```
|
||||
|
||||
⚠️ **重要**：`max_open_trades=1` 和 `stake_amount="unlimited"` 是收益率数字的关键——它们确保每笔交易使用全部资金，实现完整复利。如果改成 `max_open_trades=3`，收益会被人为压低到约 1/3。
|
||||
|
||||
---
|
||||
|
||||
## 13. 演进历史
|
||||
|
||||
策略从 v0.1 到 v2.1 经历了多轮迭代：
|
||||
|
||||
```
|
||||
v0.x 系列（指标策略）
|
||||
├── 使用 EMA/ATR 等衍生指标
|
||||
└── ❌ 已弃用：用户要求纯价格行为学
|
||||
|
||||
v1.0 ~ v1.5（探索期）
|
||||
├── v1.0: 5M TF, spot → 噪音太多
|
||||
├── v1.1: 1H TF, futures, 做空 → +61%
|
||||
├── v1.2: Entry Candle 止损 → bug
|
||||
├── v1.3: ATR 动态止损 → -63.72%
|
||||
├── v1.4: 回归价格结构止损 → +140.71% ✅
|
||||
└── v1.5: 参数调优 → +140.83%
|
||||
|
||||
v1.6（旧最优基线）
|
||||
├── 入场质量过滤：冷却期 + 活支撑/阻力
|
||||
├── ETH: +3659%, PF=3.78
|
||||
└── ✅ 证明"入场质量 > 止损优化"
|
||||
|
||||
v1.7 ~ v1.9（止损优化尝试 — 全部失败）
|
||||
├── v1.7: 5%缓冲 → 单笔亏损过大
|
||||
├── v1.8: 2%缓冲 → 总收益下降45%
|
||||
└── v1.9: 结构变化检测止损 → 收益下降56%
|
||||
|
||||
v2.0 ~ v2.1（趋势过滤）
|
||||
├── v2.0: B1入场延迟确认 → 方向正确但降频严重
|
||||
└── v2.1: D1趋势强度过滤 → ⭐ 当前最优
|
||||
```
|
||||
|
||||
### 关键教训
|
||||
|
||||
1. **止损缓冲不是越大越好** — 小缓冲 + trailing stop > 大缓冲
|
||||
2. **入场质量 > 止损优化** — 减少需要止损的交易是最佳路径
|
||||
3. **二元过滤器容易误杀** — 需要更精细的评分机制
|
||||
4. **所有优化必须先回测** — v1.7/v1.8 的失败证明了这一点
|
||||
|
||||
---
|
||||
|
||||
## 14. 已知局限
|
||||
|
||||
### 14.1 震荡市表现差
|
||||
|
||||
策略在趋势明确时表现优秀，但在震荡市中连续止损。这是因为：
|
||||
- 震荡市没有持续的 HH/HL 或 LH/LL
|
||||
- S/R 位频繁被穿越
|
||||
- K线反转信号在震荡市中可靠性低
|
||||
|
||||
**当前对策**：v2.1 的趋势强度过滤可以排除部分震荡市信号，但不能完全解决。
|
||||
|
||||
**未来方向**：市场状态分类器（趋势 vs 震荡），在震荡市自动降低交易频率或暂停。
|
||||
|
||||
### 14.2 胜率偏低（~35%）
|
||||
|
||||
65% 的交易以止损结束。这在价格行为策略中是正常的，但对交易者心态是巨大考验。连续8笔亏损跨越35天，需要极强的纪律性。
|
||||
|
||||
### 14.3 回撤偏大（~30-40%）
|
||||
|
||||
最大回撤可能达到 30-40%。10万本金可能暂时缩水到6万。
|
||||
|
||||
### 14.4 BTC 表现远弱于 ETH
|
||||
|
||||
BTC 的收益率（+575%）远低于 ETH（+4366%），差了近7倍。这可能与 ETH 波动性更大、趋势更明确有关。
|
||||
|
||||
### 14.5 未经过实盘验证
|
||||
|
||||
所有数据来自回测。实盘中可能面临：
|
||||
- 滑点
|
||||
- 网络延迟
|
||||
- 交易所 API 限制
|
||||
- 极端行情下的流动性问题
|
||||
|
||||
---
|
||||
|
||||
## 15. 部署信息
|
||||
|
||||
### 15.1 当前运行状态
|
||||
|
||||
| 项目 | 详情 |
|
||||
|:------|:------|
|
||||
| 服务器 | 腾讯云东京 43.163.225.30 |
|
||||
| 系统 | Ubuntu 24.04.3 LTS |
|
||||
| freqtrade | 2025.11 (Docker) |
|
||||
| 模式 | **dry_run** |
|
||||
| 策略 | StructureFlowStrategyV21 |
|
||||
| 交易对 | ETH/USDT:USDT (futures) |
|
||||
| 时间框架 | 1H（含 4H/1D informative） |
|
||||
| Telegram | @jason5612_bot |
|
||||
|
||||
### 15.2 服务器目录结构
|
||||
|
||||
```
|
||||
~/freqtrade/
|
||||
├── docker-compose.yml # Docker 编排配置
|
||||
└── user_data/
|
||||
├── config.json # 主配置（含 Telegram）
|
||||
├── config.pairlist.json # 交易对白名单
|
||||
├── strategies/
|
||||
│ └── structure_flow_strategy_v2_1.py
|
||||
├── data/binance/ # K线数据
|
||||
├── logs/ # 运行日志
|
||||
└── tradesv3.sqlite # 交易数据库
|
||||
```
|
||||
|
||||
### 15.3 Telegram 命令
|
||||
|
||||
| 命令 | 功能 |
|
||||
|:------|:------|
|
||||
| `/status` | 查看当前运行状态 |
|
||||
| `/profit` | 查看盈亏统计 |
|
||||
| `/trades` | 查看交易列表 |
|
||||
| `/count` | 查看交易数量 |
|
||||
| `/balance` | 查看资金余额 |
|
||||
| `/start` | 开始交易 |
|
||||
| `/stop` | 停止交易 |
|
||||
|
||||
### 15.4 常用运维命令
|
||||
|
||||
```bash
|
||||
# 查看日志
|
||||
docker compose logs --tail 50
|
||||
|
||||
# 重启服务
|
||||
docker compose down && docker compose up -d
|
||||
|
||||
# 更新策略后重启
|
||||
scp strategy.py ubuntu@43.163.225.30:~/freqtrade/user_data/strategies/
|
||||
ssh ubuntu@43.163.225.30 "cd ~/freqtrade && docker compose restart"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 附录 A：策略完整入场条件速查
|
||||
|
||||
### LONG（做多）
|
||||
|
||||
```
|
||||
trend_up_1d = True # D1 上升趋势
|
||||
in_demand_4h = True # 价格在4H需求区（下方35%）
|
||||
bullish_signal = True # 1H 出现看涨 Pin Bar 或 Engulfing
|
||||
stop_dist ≤ 5% # 入场到支撑位 ≤ 5%
|
||||
stop_dist > 0.3% # 入场到支撑位 > 0.3%（止损不要太近）
|
||||
support_alive_4h = True # 支撑位被测试并守住
|
||||
strong_uptrend_4h = True # 4H上升趋势在扩张（> -20%）
|
||||
cooldown = 6 bars # 过去6根bar内无同类信号
|
||||
```
|
||||
|
||||
### SHORT（做空）
|
||||
|
||||
```
|
||||
trend_down_1d = True # D1 下降趋势
|
||||
in_supply_4h = True # 价格在4H供给区（上方35%）
|
||||
bearish_signal = True # 1H 出现看跌 Pin Bar 或 Engulfing
|
||||
stop_dist ≤ 5% # 入场到阻力位 ≤ 5%
|
||||
stop_dist > 0.3% # 入场到阻力位 > 0.3%
|
||||
resistance_alive_4h = True # 阻力位被测试并守住
|
||||
strong_downtrend_4h = True # 4H下降趋势在扩张（> -20%）
|
||||
cooldown = 6 bars # 过去6根bar内无同类信号
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
> **最后更新**: 2026-06-08 | **作者**: 用户 + WorkBuddy AI
|
||||
>
|
||||
> 本文档将随策略迭代持续更新。
|
||||
531
docs/v1.6_strategy_doc.md
Normal file
531
docs/v1.6_strategy_doc.md
Normal file
@ -0,0 +1,531 @@
|
||||
# StructureFlowStrategy v1.6 — 完整策略解析文档
|
||||
|
||||
> 写给自己看的版本，在下一次优化开始前彻底搞清楚这 436 行代码在做什么。
|
||||
|
||||
---
|
||||
|
||||
## 一、策略大纲
|
||||
|
||||
这是一个**纯价格行为策略**，零指标（没有 MACD、RSI、布林带……），只用 K 线结构做决策。
|
||||
|
||||
核心思想只有一句话：
|
||||
|
||||
> **在大趋势方向上，找到有效的支撑/阻力位，等价格出现反转形态信号时入场，让市场结构自动追踪止损。**
|
||||
|
||||
三层时间框架，各司其职：
|
||||
|
||||
```
|
||||
D1（日线） → 判断方向：现在是涨势还是跌势？
|
||||
4H（四小时）→ 找位置：支撑/阻力在哪里？价格在需求区还是供给区？
|
||||
1H（一小时）→ 找时机：此时此刻有没有反转信号？
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 二、整体代码结构图
|
||||
|
||||
```
|
||||
StructureFlowStrategyV16
|
||||
│
|
||||
├── 工具函数（静态方法）
|
||||
│ ├── _detect_swing_points() ← 找 Swing High / Swing Low
|
||||
│ ├── _build_structure() ← 分析结构：趋势、支撑、阻力、供需区
|
||||
│ └── _detect_candle_patterns() ← 识别 K 线形态：Pin Bar、吞没
|
||||
│
|
||||
├── 指标计算（按时间框架）
|
||||
│ ├── populate_indicators_1d() ← D1：计算日线趋势方向
|
||||
│ ├── populate_indicators_4h() ← 4H：计算中期结构 + 活支撑/阻力（v1.6新增）
|
||||
│ └── populate_indicators() ← 1H：识别 K 线形态
|
||||
│
|
||||
├── 信号逻辑
|
||||
│ ├── populate_entry_trend() ← 入场条件（6个条件全满足才进）
|
||||
│ └── populate_exit_trend() ← 出场条件（D1 趋势反转时退出）
|
||||
│
|
||||
└── 动态止损
|
||||
└── custom_stoploss() ← 基于 4H Swing Point 的结构止损
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 三、基础参数说明
|
||||
|
||||
```python
|
||||
can_short = True # 允许做空（合约模式）
|
||||
stoploss = -0.15 # 兜底止损：最大允许亏损 15%（custom_stoploss 的安全边界）
|
||||
use_custom_stoploss = True # 启用自定义止损逻辑
|
||||
minimal_roi = {"0": 100} # 关闭固定止盈（100 = 永不触发）
|
||||
max_open_trades = 1 # 同时最多 1 笔交易（专注单品种）
|
||||
timeframe = "1h" # 主时间框架：1小时
|
||||
```
|
||||
|
||||
**关于 `stoploss = -0.15` 和 `custom_stoploss` 的关系：**
|
||||
|
||||
- `custom_stoploss` 是每根 K 线都会调用的动态止损逻辑
|
||||
- `-0.15` 是**硬边界**：即使 `custom_stoploss` 返回了一个更宽的止损，freqtrade 也会把它截断到 -15%
|
||||
- 这两个是"主控 + 保险"的关系，不是矛盾的
|
||||
|
||||
---
|
||||
|
||||
## 四、可优化参数
|
||||
|
||||
```python
|
||||
swing_lookback_d1 = IntParameter(8, 14, default=10, space="buy")
|
||||
# D1 识别 Swing Point 的回溯窗口：10 意味着一个高点必须是
|
||||
# 左侧 10 根日线和右侧 10 根日线里的最高点，才算 Swing High。
|
||||
# 数字越大 → Swing Point 越稀疏越重要
|
||||
# 数字越小 → Swing Point 越密集越灵敏
|
||||
|
||||
swing_lookback_h4 = IntParameter(5, 10, default=8, space="buy")
|
||||
# 4H 级别的 Swing Point 回溯窗口，逻辑同上
|
||||
|
||||
pin_bar_wick_ratio = IntParameter(50, 70, default=60, space="buy")
|
||||
# Pin Bar 识别阈值：影线占整根 K 线的比例
|
||||
# 60 表示 "影线占比 > 60% 才算 Pin Bar"
|
||||
# 数字越大 → 只认最标准的 Pin Bar（更严格）
|
||||
# 数字越小 → 稍微有点影线就算（更宽松）
|
||||
|
||||
max_stop_dist = IntParameter(20, 50, default=50, space="buy")
|
||||
# 止损距离上限：入场价到支撑位的距离不能超过 50%
|
||||
# 防止"支撑位太远、止损太大"的情况
|
||||
# 实际上 default=50 几乎不过滤（等于没有限制）
|
||||
|
||||
cooldown_bars = IntParameter(3, 12, default=6, space="buy")
|
||||
# v1.6 新增：冷却期（单位：1H K 线根数）
|
||||
# 6 表示：上一个同方向信号出现后，6 小时内不再产生新信号
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 五、工具函数详解
|
||||
|
||||
### 5.1 `_detect_swing_points()` — 找摆动高低点
|
||||
|
||||
```python
|
||||
for i in range(window, n - window):
|
||||
# 条件：第 i 根的 high > 左边 window 根的最高 AND > 右边 window 根的最高
|
||||
if high[i] > high[i-window:i].max() and high[i] > high[i+1:i+window+1].max():
|
||||
sh[i] = high[i] # 这是一个 Swing High
|
||||
# 对称的逻辑找 Swing Low
|
||||
if low[i] < low[i-window:i].min() and low[i] < low[i+1:i+window+1].min():
|
||||
sl[i] = low[i] # 这是一个 Swing Low
|
||||
```
|
||||
|
||||
**视觉理解：**
|
||||
|
||||
```
|
||||
▲ Swing High（左右各 window 根都是低点）
|
||||
/|\
|
||||
/ | \
|
||||
───────────────/ | \───────────
|
||||
```
|
||||
|
||||
**重要细节：** `range(window, n - window)` 意味着最后 `window` 根 K 线无法被识别为 Swing Point（因为需要看右边的数据）。这在实盘中意味着**最近的高低点要滞后 window 根才被确认**。
|
||||
|
||||
---
|
||||
|
||||
### 5.2 `_build_structure()` — 分析市场结构
|
||||
|
||||
这个函数是整个策略的大脑，逐根 K 线地维护 5 个状态：
|
||||
|
||||
#### （1）趋势判断
|
||||
|
||||
```python
|
||||
# 记录最近 4 个 Swing High 和 4 个 Swing Low
|
||||
sh_prices = [] # 最多保留 4 个
|
||||
sl_prices = []
|
||||
|
||||
# 趋势判断逻辑（HH/HL vs LH/LL）
|
||||
if sh_prices[-1] > sh_prices[-2] and sl_prices[-1] > sl_prices[-2]:
|
||||
trend_up = True # HH (Higher High) + HL (Higher Low) = 上升趋势
|
||||
elif sh_prices[-1] < sh_prices[-2] and sl_prices[-1] < sl_prices[-2]:
|
||||
trend_down = True # LH (Lower High) + LL (Lower Low) = 下降趋势
|
||||
else:
|
||||
# 结构不明确时，继承上一根的趋势（趋势具有惯性）
|
||||
trend_up[i] = trend_up[i-1]
|
||||
trend_down[i] = trend_down[i-1]
|
||||
```
|
||||
|
||||
**这就是价格行为学的核心：**
|
||||
- 上升趋势 = Higher High + Higher Low（HH + HL）
|
||||
- 下降趋势 = Lower High + Lower Low（LH + LL）
|
||||
|
||||
#### （2）最近支撑/阻力
|
||||
|
||||
```python
|
||||
nearest_support[i] = sl_prices[-1] # 最近的 Swing Low = 支撑
|
||||
nearest_resistance[i] = sh_prices[-1] # 最近的 Swing High = 阻力
|
||||
```
|
||||
|
||||
**注意：** 这里直接取最后一个 Swing Low 作为支撑，没有考虑位置关系（支撑可能在当前价格之上）。这是一个潜在的逻辑弱点，后续优化可以考虑。
|
||||
|
||||
#### （3）供需区划分
|
||||
|
||||
```python
|
||||
zone_range = resistance - support
|
||||
pos_pct = (close - support) / zone_range
|
||||
|
||||
in_demand_zone = pos_pct < 0.35 # 价格在 support↔resistance 区间的底部 35%
|
||||
in_supply_zone = pos_pct > 0.65 # 价格在 support↔resistance 区间的顶部 35%
|
||||
```
|
||||
|
||||
**视觉理解：**
|
||||
|
||||
```
|
||||
┌─────────────────────────────────┐
|
||||
│ resistance ────────────────── │
|
||||
│ 供给区（Supply Zone, >65%） │ ← 做空区域
|
||||
│ ─────────────────────────── 65%│
|
||||
│ │
|
||||
│ 中间区（35%~65%） │ ← 不入场
|
||||
│ │
|
||||
│ ─────────────────────────── 35%│
|
||||
│ 需求区（Demand Zone, <35%） │ ← 做多区域
|
||||
│ support ──────────────────── │
|
||||
└─────────────────────────────────┘
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
### 5.3 `_detect_candle_patterns()` — K 线形态识别
|
||||
|
||||
#### Pin Bar（Pin 柱/别针 K 线）
|
||||
|
||||
```python
|
||||
body = abs(close - open_) # 实体大小
|
||||
total_range = high - low # 整根 K 线的范围
|
||||
upper_wick = high - max(close, open_) # 上影线长度
|
||||
lower_wick = min(close, open_) - low # 下影线长度
|
||||
|
||||
is_pin = (upper_wick + lower_wick) / total_range > pin_bar_wick_ratio
|
||||
# 当影线占比 > 60%，才算 Pin Bar
|
||||
|
||||
# 看涨 Pin Bar：阳线 + 下影线 > 上影线（价格被撑住了）
|
||||
bullish_pin = is_pin & (close > open_) & (lower_wick > upper_wick)
|
||||
|
||||
# 看跌 Pin Bar：阴线 + 上影线 > 下影线（价格被压下来了）
|
||||
bearish_pin = is_pin & (close < open_) & (upper_wick > lower_wick)
|
||||
```
|
||||
|
||||
**视觉理解：**
|
||||
|
||||
```
|
||||
看涨 Pin Bar 看跌 Pin Bar
|
||||
│ ██████
|
||||
│ │
|
||||
███ │
|
||||
│ │
|
||||
███████ ███████
|
||||
│
|
||||
███████
|
||||
```
|
||||
|
||||
#### 吞没形态（Engulfing）
|
||||
|
||||
```python
|
||||
# 看涨吞没：阳线，收盘 > 前根开盘，开盘 < 前根收盘
|
||||
bullish_engulf = (close > prev_open) & (open_ < prev_close) & (close > open_)
|
||||
|
||||
# 看跌吞没：阴线，收盘 < 前根开盘，开盘 > 前根收盘
|
||||
bearish_engulf = (close < prev_open) & (open_ > prev_close) & (close < open_)
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 六、时间框架指标详解
|
||||
|
||||
### 6.1 D1 日线（`populate_indicators_1d`）
|
||||
|
||||
**只做一件事：确定宏观趋势方向**
|
||||
|
||||
- `trend_up_1d = True` → 日线是上升趋势（HH + HL）
|
||||
- `trend_down_1d = True` → 日线是下降趋势（LH + LL）
|
||||
|
||||
回溯窗口 `swing_lookback_d1 = 10`，意味着确认一个日线 Swing Point 需要左右各 10 天，合计 20 天。这是有意为之——**日线趋势应该稳定，不应该频繁切换**。
|
||||
|
||||
---
|
||||
|
||||
### 6.2 4H 四小时（`populate_indicators_4h`）
|
||||
|
||||
**做五件事：**
|
||||
|
||||
1. `trend_up_4h` / `trend_down_4h`：4H 中期趋势（与 D1 配合，双重确认）
|
||||
2. `support_4h`：最近的 4H Swing Low（止损基准点）
|
||||
3. `resistance_4h`：最近的 4H Swing High（做空止损基准点）
|
||||
4. `in_demand_4h`：当前 1H 价格是否在 4H 需求区（做多区域）
|
||||
5. `in_supply_4h`：当前 1H 价格是否在 4H 供给区（做空区域）
|
||||
|
||||
**v1.6 新增：活支撑/阻力判断**
|
||||
|
||||
```python
|
||||
# 支撑"活"的判断：
|
||||
# 条件1：最近某根 4H K 线的 low，触及 support ± 0.5% 的范围
|
||||
touched_support = (low <= support * 1.005) & (low >= support * 0.995)
|
||||
|
||||
# 条件2：那根 K 线的收盘价在 support 之上（撑住了，没跌穿）
|
||||
held_support = close > support
|
||||
|
||||
# 两个条件都满足，才算"一次有效测试"
|
||||
support_tested_and_held = touched_support & held_support
|
||||
|
||||
# 在最近 3 根 4H K 线内，至少发生过一次有效测试 → 支撑是"活"的
|
||||
dataframe["support_alive"] = support_tested_and_held.rolling(3, min_periods=1).max() > 0
|
||||
```
|
||||
|
||||
**为什么要"活支撑"？**
|
||||
|
||||
"死支撑"是指那些很久以前的 Swing Low，现在价格距离它很远，支撑位已经失效。例如：
|
||||
|
||||
```
|
||||
3个月前的 Swing Low = 1500 美元（支撑）
|
||||
但价格从那之后一直在 2000-2500 之间波动，从未回测过 1500
|
||||
→ 1500 的支撑已经是"死"的，在那里做多没有意义
|
||||
```
|
||||
|
||||
v1.6 要求：支撑必须是**最近 3 根 4H K 线内有价格来测试过的**，才算有效。
|
||||
|
||||
**"活支撑" 的问题（客观评价）：**
|
||||
- `rolling(3)` = 最近 12 小时（3 × 4H）内测试过
|
||||
- 这个窗口是否合适？12 小时可能太短，也可能太长
|
||||
- 是未来优化的方向之一
|
||||
|
||||
---
|
||||
|
||||
### 6.3 1H 一小时（`populate_indicators`）
|
||||
|
||||
**只做一件事：识别 K 线反转形态**
|
||||
|
||||
```python
|
||||
dataframe["bullish_signal"] = bullish_pin | bullish_engulf # 两种看涨形态满足任一
|
||||
dataframe["bearish_signal"] = bearish_pin | bearish_engulf # 两种看跌形态满足任一
|
||||
```
|
||||
|
||||
**NaN 安全处理：**
|
||||
由于不同时间框架的数据合并（D1、4H → 1H）会产生空值，这里统一对布尔列做 `fillna(False)`，防止 `True & NaN = NaN` 导致信号丢失。
|
||||
|
||||
---
|
||||
|
||||
## 七、入场逻辑详解
|
||||
|
||||
### 做多条件（6 个条件全部满足）
|
||||
|
||||
```python
|
||||
long_base = (
|
||||
dataframe["trend_up_1d"] # 条件1：日线是上升趋势
|
||||
& dataframe["in_demand_4h"] # 条件2：当前价格在 4H 需求区（支撑附近）
|
||||
& dataframe["bullish_signal"] # 条件3：1H 出现看涨 K 线形态
|
||||
& (long_stop_dist <= max_dist) # 条件4：止损距离 ≤ 50%（几乎不过滤）
|
||||
& (long_stop_dist > 0.003) # 条件5：止损距离 > 0.3%（防止支撑=开盘价）
|
||||
)
|
||||
|
||||
long_base = long_base & dataframe["support_alive_4h"] # 条件6：支撑是"活"的 [v1.6]
|
||||
|
||||
long_recent = long_base.rolling(cooldown).max().shift(1) == 0 # 冷却期过滤 [v1.6]
|
||||
long_conditions = long_base & long_recent
|
||||
```
|
||||
|
||||
**条件5 的意义：** `long_stop_dist > 0.003` 防止"支撑位和开盘价几乎相同"的情况，那样止损太近，任何正常波动都会触发。
|
||||
|
||||
### 冷却期的工作原理
|
||||
|
||||
```python
|
||||
# long_base 是"条件1-6都满足"的信号（不含冷却）
|
||||
# rolling(6).max() → 过去6根1H bar内，是否有过满足条件的信号（1=有，0=没有）
|
||||
# .shift(1) → 往后移一格，避免当前这根K线跟自己比较
|
||||
# == 0 → 只有"过去6小时内没有信号"时，才允许当前入场
|
||||
|
||||
long_recent = long_base.rolling(cooldown, min_periods=1).max().shift(1) == 0
|
||||
```
|
||||
|
||||
**视觉理解（cooldown=6）：**
|
||||
|
||||
```
|
||||
时间 → 1h 2h 3h 4h 5h 6h 7h 8h 9h 10h
|
||||
信号 ✓ ✓
|
||||
冷却期 ←── 6小时冷却 ──→
|
||||
允许入场 ✓ ✓ （第10小时重新可以入场）
|
||||
```
|
||||
|
||||
### 做空条件（对称）
|
||||
|
||||
```python
|
||||
short_base = (
|
||||
dataframe["trend_down_1d"] # 条件1：日线是下降趋势
|
||||
& dataframe["in_supply_4h"] # 条件2：价格在 4H 供给区（阻力附近）
|
||||
& dataframe["bearish_signal"] # 条件3：1H 出现看跌 K 线形态
|
||||
& (short_stop_dist <= max_dist) # 条件4：止损距离合理
|
||||
& (short_stop_dist > 0.003) # 条件5：止损距离不为零
|
||||
)
|
||||
short_base = short_base & dataframe["resistance_alive_4h"] # 条件6：阻力是"活"的
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 八、出场逻辑详解
|
||||
|
||||
```python
|
||||
exit_long = ~dataframe["trend_up_1d"].fillna(True) # 当 D1 趋势不再上升时平多
|
||||
exit_short = dataframe["trend_up_1d"].fillna(False) # 当 D1 趋势转为上升时平空
|
||||
```
|
||||
|
||||
**这个逻辑非常简单，也非常粗糙：**
|
||||
|
||||
- 只有 D1 趋势反转时才触发出场信号
|
||||
- D1 趋势需要明确的 HH+HL 或 LH+LL 才会切换，这需要较长时间
|
||||
- 结果：大多数交易不会触发 exit_signal，而是由止损（stop_loss 或 trailing_stop_loss）退出
|
||||
|
||||
这也是为什么回测中 `exit_signal` 只有 7-9 笔，而止损退出有 70-100 笔。
|
||||
|
||||
---
|
||||
|
||||
## 九、动态止损详解（最核心的部分）
|
||||
|
||||
```python
|
||||
def custom_stoploss(self, pair, trade, current_time, current_rate, current_profit, ...):
|
||||
last = dataframe.iloc[-1] # 取最新的 4H 数据
|
||||
|
||||
if not trade.is_short: # 多单
|
||||
support = last["support_4h"] # 最近的 4H Swing Low
|
||||
sl_price = support * 0.999 # 止损价 = 支撑位下方 0.1%
|
||||
sl_ratio = (sl_price / current_rate) - 1.0 # 转换为相对于当前价格的比率
|
||||
return max(sl_ratio, -0.15) # 不超过 -15% 的硬边界
|
||||
else: # 空单
|
||||
resistance = last["resistance_4h"] # 最近的 4H Swing High
|
||||
sl_price = resistance * 1.001 # 止损价 = 阻力位上方 0.1%
|
||||
sl_ratio = 1.0 - (sl_price / current_rate)
|
||||
return min(sl_ratio, 0.15) # 不超过 +15% 的硬边界
|
||||
```
|
||||
|
||||
### 为什么这会产生"自然追踪止损效果"？
|
||||
|
||||
关键在于 `support_4h` 是**动态更新**的：
|
||||
|
||||
```
|
||||
t=0 入场，support_4h = 1800（止损在 1798.2）
|
||||
价格从 1850 开始上涨
|
||||
t=5h 价格涨到 2000，出现新 Swing Low = 1900
|
||||
→ support_4h 更新为 1900
|
||||
→ 止损自动上移到 1898.1
|
||||
t=10h 价格涨到 2200，新 Swing Low = 2050
|
||||
→ support_4h 更新为 2050
|
||||
→ 止损上移到 2047.9
|
||||
```
|
||||
|
||||
这就是为什么 `trailing_stop_loss` 出现那么多：每当价格创新高、产生新的 Swing Low，止损就自动追上去，锁住利润。
|
||||
|
||||
### 止损的数学计算示例
|
||||
|
||||
假设：
|
||||
- 当前价格 = 2000 USDT
|
||||
- support_4h = 1900 USDT
|
||||
|
||||
```python
|
||||
sl_price = 1900 * 0.999 = 1898.1
|
||||
sl_ratio = (1898.1 / 2000) - 1.0 = -0.0595（约 -5.95%）
|
||||
```
|
||||
|
||||
即：止损触发点在当前价格的 5.95% 下方。
|
||||
|
||||
---
|
||||
|
||||
## 十、退出原因分类（理解 freqtrade 的逻辑）
|
||||
|
||||
| 退出原因 | 触发条件 | v1.6 ETH 全周期 |
|
||||
|----------|----------|-----------------|
|
||||
| `stop_loss` | 价格触及 `custom_stoploss` 返回的价格，且 current_profit < 0 | 73 笔，0% 胜率 |
|
||||
| `trailing_stop_loss` | 价格从高点回落，触及之前计算的止损价，且有过盈利 | 71 笔，67.6% 胜率 |
|
||||
| `exit_signal` | `populate_exit_trend` 返回 exit=1 | 7 笔，71.4% 胜率 |
|
||||
| `force_exit` | 回测结束强制平仓 | 1 笔，100% |
|
||||
|
||||
**`stop_loss` 和 `trailing_stop_loss` 的本质区别：**
|
||||
|
||||
- `stop_loss`：交易从入场到退出**从未产生过盈利**。价格一直往不利方向走，直到触发止损。
|
||||
- `trailing_stop_loss`：交易曾经**有过盈利**，后来价格回落，触及追踪止损位退出。
|
||||
|
||||
**这说明了什么？**
|
||||
|
||||
73 笔 `stop_loss` = 73 笔入场后价格立即向不利方向走的交易。
|
||||
|
||||
这些交易**入场位是错的**，或者**在错误的时机做了正确的方向**。这是 v1.6 最核心的问题，也是下一步优化的主攻方向。
|
||||
|
||||
---
|
||||
|
||||
## 十一、已知问题和局限性
|
||||
|
||||
### 问题 1：`stop_loss` 胜率 0%（最严重）
|
||||
- 表现：73 笔 stop_loss 退出，全部亏损
|
||||
- 根本原因：入场后价格立即反向，说明入场位质量不高
|
||||
- 可能方向：更严格的入场过滤、更好的时机判断
|
||||
|
||||
### 问题 2：支撑位识别的逻辑漏洞
|
||||
- `nearest_support = sl_prices[-1]` 直接取最近的 Swing Low，可能在当前价格之上
|
||||
- 例如：价格在 2000，但最近 Swing Low 是 2100（价格刚从 2100 跌下来），这时支撑 = 2100 > 2000，止损价在当前价格上方，逻辑矛盾
|
||||
- `long_stop_dist > 0.003` 一定程度上过滤了这种情况，但不完全
|
||||
|
||||
### 问题 3：出场逻辑太粗糙
|
||||
- 只看 D1 趋势是否反转
|
||||
- D1 趋势切换很慢，大多数交易靠止损退出，而非出场信号
|
||||
- 可以考虑增加 4H 结构破坏作为出场信号
|
||||
|
||||
### 问题 4：市场环境依赖
|
||||
- 趋势市（2023-2024）：表现优异
|
||||
- 震荡市（2025 YTD BTC）：表现差
|
||||
- 没有"市场环境过滤器"，在震荡市会频繁入场并被止损
|
||||
|
||||
### 问题 5：`fillna(False)` 的 FutureWarning
|
||||
- 技术性问题，不影响结果，但需要修复（pandas 未来版本会报错）
|
||||
|
||||
---
|
||||
|
||||
## 十二、数据流的完整路径
|
||||
|
||||
一笔做多交易的完整生命周期：
|
||||
|
||||
```
|
||||
1. D1 数据 → _detect_swing_points() → _build_structure()
|
||||
→ 输出 trend_up_1d = True （日线上升趋势确认）
|
||||
|
||||
2. 4H 数据 → _detect_swing_points() → _build_structure()
|
||||
→ 输出 support_4h = 1900, resistance_4h = 2100
|
||||
→ 输出 in_demand_4h = True （价格在需求区）
|
||||
→ v1.6: 检查 support_alive_4h = True （支撑最近被测试过）
|
||||
|
||||
3. 1H 数据 → _detect_candle_patterns()
|
||||
→ 输出 bullish_signal = True （出现看涨 Pin Bar 或吞没）
|
||||
|
||||
4. populate_entry_trend() 汇总所有条件
|
||||
→ 全部满足 + 冷却期过了 → enter_long = 1 → 下单做多
|
||||
|
||||
5. 持仓期间，每根 1H K 线：
|
||||
→ custom_stoploss() 被调用
|
||||
→ 读取最新 support_4h
|
||||
→ 计算止损价 = support_4h * 0.999
|
||||
→ 止损位随 support_4h 上移（追踪效果）
|
||||
|
||||
6. 退出：
|
||||
→ 价格创新高后回落触及止损 → trailing_stop_loss 退出
|
||||
→ 或价格入场后就下跌 → stop_loss 退出
|
||||
→ 或 D1 趋势反转 → exit_signal 退出
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 十三、优化的可能方向（供后续参考）
|
||||
|
||||
基于以上分析，以下是最有价值的优化方向（按优先级排序）：
|
||||
|
||||
**优先级 A：解决 stop_loss 0% 胜率**
|
||||
1. 分析 73 笔 stop_loss 交易的特征（何时入场？市场处于什么状态？）
|
||||
2. 增加入场质量过滤（例如：要求 4H 也是上升趋势、要求成交量确认）
|
||||
3. 改进 Swing Point 识别（例如：要求支撑位被测试 2 次以上）
|
||||
|
||||
**优先级 B：增加市场环境过滤**
|
||||
1. 识别"震荡市"（例如：最近 N 根 4H bar 内，没有明确的 HH/HL 序列）
|
||||
2. 在震荡市中停止交易，只在趋势市中工作
|
||||
|
||||
**优先级 C：改进出场逻辑**
|
||||
1. 将出场信号细化到 4H 级别（4H 结构破坏时退出，而不等 D1 反转）
|
||||
2. 可以大幅减少 stop_loss 次数，让更多交易变成 exit_signal 或 trailing_stop_loss
|
||||
|
||||
---
|
||||
|
||||
*文档基于 `structure_flow_strategy_v1_6.py` 代码，v1.6 版本（2026-06-07）*
|
||||
Reference in New Issue
Block a user