v3.2 (Swing): 多TF震荡确认 + 供需区动态评分

strategy.py

@@ -1,21 +1,28 @@
 """
-Structure Flow Swing Strategy v3.1
+Structure Flow Swing Strategy v3.2
 ==================================
-波段交易策略 — 基于4H震荡区间，保守参数 v2
+波段交易策略 — 基于4H震荡区间，v3.1优化版
 
-v3.1 改动（基于v3.0诊断结果）：
+v3.2 改动（基于v3.1诊断结果 — 三大市场感知不足）：
-  1. 双边测试 AND→OR：在10根K线内测试过支撑 OR 阻力即可（不需两者都测过）
+  1. D1趋势强度过滤：D1处于强趋势时拒绝入场，防假区间陷阱
-  2. 区间稳定性 15%→25%：放宽波动容忍度
+     - 计算 D1 EMA20/EMA50 间距作为趋势强度指标
-  3. 入场范围 2%→3%：增加候选信号密度
+     - 趋势强度超过阈值 → 不交易（即使4H出现区间形态）
-  4. 冷却期 3根→1根：减少过渡过滤
+  2. 区间质量评分：从二分法升级为多维度评分
+     - 边界测试次数（测试越多越可靠）
+     - 区间持续时长（越长越成熟）
+     - 区间宽度适配度（3-8%最优）
+     - 总分>=阈值才入场
+  3. 主动退出机制：确认转趋势后提前离场
+     - 3根连续K线收盘在入场时区间外 → 结构破坏
+     - 不等止损，主动离场（仅在损失<2%时）
+     - 避免浮盈变亏损
 
-保留：纯震荡定位、ATR×1.5止损、区间70%止盈、D1趋势过滤
+保留：纯震荡定位、ATR×1.5止损、区间70%止盈、OR双边测试、冷却期1根
-
-预期：年交易量从9笔 → 50-80笔（约1-2单/周）
 
 版本历史：
   v3.0 (2026-06-10): 初版，基于冯总波段交易新思路
-  v3.1 (2026-06-10): 降低条件门槛，提升交易频率
+  v3.1 (2026-06-10): 降低条件门槛，AND→OR等4项
+  v3.2 (2026-06-10): 三大市场感知改进
 """
 
 from datetime import datetime
@@ -26,10 +33,10 @@ from freqtrade.strategy import IStrategy, IntParameter, informative
 from freqtrade.persistence import Trade
 
 
-class StructureFlowSwingV31(IStrategy):
+class StructureFlowSwingV32(IStrategy):
     """
-    Structure Flow Swing Strategy v3.1
+    Structure Flow Swing Strategy v3.2
-    4H震荡区间波段交易 — 放宽震荡判定
+    4H震荡区间波段交易 — 市场感知增强版
     """
 
     can_short = True
@@ -40,17 +47,22 @@ class StructureFlowSwingV31(IStrategy):
     timeframe = "4h"
 
     # =====================
-    # 可优化参数（放宽后默认值）
+    # 核心参数（沿用v3.1默认值）
     # =====================
     swing_lookback = IntParameter(4, 8, default=5, space="buy")
-    zone_stability_threshold = IntParameter(15, 40, default=25, space="buy")  # v3.1: 15→25↑
+    zone_stability_threshold = IntParameter(15, 40, default=25, space="buy")
-    entry_zone_pct = IntParameter(1, 5, default=3, space="buy")               # v3.1: 2→3↑
+    entry_zone_pct = IntParameter(1, 5, default=3, space="buy")
     atr_stop_mult = IntParameter(10, 25, default=15, space="buy")
     take_profit_pct = IntParameter(50, 80, default=70, space="sell")
 
+    # v3.2 新增参数
+    d1_trend_strength_max = IntParameter(6, 15, default=10, space="buy")  # D1趋势强度上限%，默认10%（极端趋势才触发）
+    zone_quality_min = IntParameter(20, 60, default=30, space="buy")      # 区间质量最低分，默认30
+
     # 固定参数
     zone_touch_lookback = 10
     breakout_bars = 2
+    early_exit_bars = 3      # v3.2新增：连续N根在区间外触发主动退出
 
     # =====================
     # 工具：Swing Point 检测
@@ -73,7 +85,7 @@ class StructureFlowSwingV31(IStrategy):
         return sh, sl
 
     # =====================
-    # 工具：区间震荡检测
+    # 工具：区间震荡检测（增强版：加入质量评分数据）
     # =====================
 
     def _detect_range(
@@ -89,10 +101,14 @@ class StructureFlowSwingV31(IStrategy):
         support_arr = np.full(n, np.nan)
         resistance_arr = np.full(n, np.nan)
         zone_width_arr = np.full(n, np.nan)
+        touch_count_arr = np.full(n, 0)        # v3.2新增
 
         sh_prices = []
         sl_prices = []
 
+        in_range = False
+        touch_count = 0
+
         for i in range(n):
             if pd.notna(sh.iloc[i]):
                 sh_prices.append(sh.iloc[i])
@@ -104,12 +120,19 @@ class StructureFlowSwingV31(IStrategy):
                     sl_prices.pop(0)
 
             if len(sh_prices) < 3 or len(sl_prices) < 3:
+                # 不在区间中
+                if in_range:
+                    in_range = False
+                    touch_count = 0
                 continue
 
             current_sh = sh_prices[-1]
             current_sl = sl_prices[-1]
 
             if current_sh <= current_sl:
+                if in_range:
+                    in_range = False
+                    touch_count = 0
                 continue
 
             zone_width = (current_sh - current_sl) / current_sl
@@ -137,10 +160,12 @@ class StructureFlowSwingV31(IStrategy):
                 is_stable = False
 
             if not is_stable:
+                if in_range:
+                    in_range = False
+                    touch_count = 0
                 continue
 
             # 条件2：价格测试过边界 — v3.1: AND→OR
-            # 只需要测试过支撑或阻力之一，不需要两者都测过
             start_idx = max(0, i - self.zone_touch_lookback)
             support_zone_upper = current_sl * 1.01
             touched_support = any(
@@ -153,8 +178,10 @@ class StructureFlowSwingV31(IStrategy):
                 for j in range(start_idx, i + 1)
             )
 
-            # v3.1: AND → OR
             if not (touched_support or touched_resistance):
+                if in_range:
+                    in_range = False
+                    touch_count = 0
                 continue
 
             # 条件3：无突破
@@ -166,15 +193,30 @@ class StructureFlowSwingV31(IStrategy):
                     break
 
             if consecutive_outside >= self.breakout_bars:
+                if in_range:
+                    in_range = False
+                    touch_count = 0
                 continue
 
+            # === 通过所有条件 → 在区间中 ===
             is_ranging[i] = True
 
+            # v3.2: 跟踪区间内的边界触碰次数（质量评分数据）
+            if not in_range:
+                in_range = True
+                touch_count = 0
+
+            c = close.iloc[i]
+            if (c <= current_sl * 1.015) or (c >= current_sh * 0.985):
+                touch_count += 1
+            touch_count_arr[i] = touch_count
+
         return DataFrame({
             "is_ranging": is_ranging,
             "support": support_arr,
             "resistance": resistance_arr,
             "zone_width": zone_width_arr,
+            "touch_count": touch_count_arr,          # v3.2新增
         }, index=high.index)
 
     # =====================
@@ -191,11 +233,12 @@ class StructureFlowSwingV31(IStrategy):
         return tr.rolling(period).mean()
 
     # ================================================================
-    # D1 信息时间框架 — 宏观趋势参考
+    # D1 信息时间框架 — v3.2: 新增趋势强度计算
     # ================================================================
 
     @informative("1d")
     def populate_indicators_1d(self, dataframe: DataFrame, metadata: dict) -> DataFrame:
+        # 原有：D1趋势方向（swing point比较）
         sh, sl = self._detect_swing_points(
             dataframe["high"], dataframe["low"], window=5
         )
@@ -213,10 +256,16 @@ class StructureFlowSwingV31(IStrategy):
 
         dataframe["d1_uptrend"] = is_uptrend
         dataframe["d1_downtrend"] = is_downtrend
+
+        # v3.2新增：D1趋势强度 = EMA20与EMA50的偏离程度
+        ema_20 = dataframe["close"].ewm(span=20, adjust=False).mean()
+        ema_50 = dataframe["close"].ewm(span=50, adjust=False).mean()
+        dataframe["trend_strength"] = abs(ema_20 - ema_50) / ema_50
+
         return dataframe
 
     # ================================================================
-    # 主时间框架 — 4H 指标
+    # 主时间框架 — 4H 指标（v3.2: 新增区间质量评分）
     # ================================================================
 
     def populate_indicators(self, dataframe: DataFrame, metadata: dict) -> DataFrame:
@@ -230,6 +279,7 @@ class StructureFlowSwingV31(IStrategy):
         dataframe["range_support"] = range_info["support"]
         dataframe["range_resistance"] = range_info["resistance"]
         dataframe["zone_width_pct"] = range_info["zone_width"]
+        dataframe["range_touch_count"] = range_info["touch_count"]
 
         dataframe["atr"] = self._calc_atr(dataframe["high"], dataframe["low"], dataframe["close"], 14)
 
@@ -253,14 +303,66 @@ class StructureFlowSwingV31(IStrategy):
             np.nan,
         )
 
+        # ── v3.2新增：区间质量评分 ──
+        self._compute_zone_quality(dataframe)
+
+        # ── v3.2新增：区间连续计数 ──
+        is_ranging_int = dataframe["is_ranging"].astype(int)
+        consecutive = np.zeros(len(dataframe), dtype=int)
+        for i in range(1, len(dataframe)):
+            if is_ranging_int.iloc[i] and is_ranging_int.iloc[i-1]:
+                consecutive[i] = consecutive[i-1] + 1
+            elif is_ranging_int.iloc[i]:
+                consecutive[i] = 1
+        dataframe["range_consecutive"] = consecutive
+
         for col in ["is_ranging", "zone_position", "dist_to_support", "dist_to_resistance"]:
             if col in dataframe.columns:
                 dataframe[col] = dataframe[col].fillna(False if col == "is_ranging" else 999)
 
         return dataframe
 
+    def _compute_zone_quality(self, dataframe: DataFrame) -> None:
+        """
+        v3.2新增：区间质量三因子评分
+        - 边界测试次数（0-45分）：0→15, 1→20, 2→32, 3+→45
+        - 区间持续时长（0-30分）：<5→0, 5-9→12, 10-19→22, 20+→30
+        - 区间宽度适配（0-25分）：3-8%→25, 2-3%→15, 8-12%→15, 其他→0
+        满分100，合格线默认30
+        """
+        touch_count = dataframe["range_touch_count"].fillna(0).values
+        zone_width = dataframe["zone_width_pct"].fillna(0).values
+        is_ranging = dataframe["is_ranging"].values
+
+        quality = np.zeros(len(dataframe))
+
+        # 因子1：边界测试次数（放宽：0次触碰也有基础分）
+        quality += np.where(
+            touch_count >= 3, 45,
+            np.where(touch_count >= 2, 32,
+                     np.where(touch_count >= 1, 20, 15))
+        )
+
+        # 因子2：区间持续时长（用连续计数表示暂存，后续由 populate_indicators 补充）
+        # 这里先按最少给分，populate_indicators 中会基于 range_consecutive 二次修正
+        # 实际上 touche_count > 0 就意味着至少有一些持续性
+
+        # 因子3：区间宽度适配度
+        quality += np.where(
+            (zone_width >= 0.03) & (zone_width <= 0.08), 25,
+            np.where(
+                ((zone_width >= 0.02) & (zone_width < 0.03)) |
+                ((zone_width > 0.08) & (zone_width <= 0.12)), 15, 0
+            )
+        )
+
+        # 只在区间内有效
+        quality = np.where(is_ranging, quality, 0)
+
+        dataframe["zone_quality_base"] = quality
+
     # ================================================================
-    # 入场信号 — v3.1: 冷却期 3→1
+    # 入场信号 — v3.2: D1趋势强度 + 区间质量过滤 + 持续时间因子
     # ================================================================
 
     def populate_entry_trend(self, dataframe: DataFrame, metadata: dict) -> DataFrame:
@@ -268,22 +370,54 @@ class StructureFlowSwingV31(IStrategy):
 
         d1_downtrend_col = "d1_downtrend_1d"
         d1_uptrend_col = "d1_uptrend_1d"
+        d1_strength_col = "trend_strength_1d"  # v3.2新增
 
-        for col in ["is_ranging", d1_uptrend_col, d1_downtrend_col]:
+        for col in ["is_ranging", d1_uptrend_col, d1_downtrend_col, d1_strength_col]:
             if col in dataframe.columns:
                 dataframe[col] = dataframe[col].fillna(False)
             else:
                 dataframe[col] = False
 
+        # ── v3.2: 计算完整区间质量评分（加入持续性因子） ──
+        range_consec = dataframe.get("range_consecutive", pd.Series(0, index=dataframe.index))
+        quality_base = dataframe.get("zone_quality_base", pd.Series(0, index=dataframe.index))
+
+        # 持续性因子：<5→0, 5-9→12, 10-19→22, 20+→30
+        duration_score = np.where(
+            range_consec >= 20, 30,
+            np.where(range_consec >= 10, 22,
+                     np.where(range_consec >= 5, 12, 0))
+        )
+
+        # 完整质量分 = 基础分（测试+宽度，max=70）+ 持续性分（max=30）
+        dataframe["zone_quality"] = quality_base + duration_score
+        dataframe["zone_quality"] = np.where(dataframe["is_ranging"], dataframe["zone_quality"], 0)
+
+        # ── v3.2: D1趋势强度过滤（方向感知） ──
+        # 逻辑：只有在极端趋势中，同向的4H区间才有"假区间"风险
+        #   - 做多：D1处于极端上升趋势 → 回调可能很深 → 不进场
+        #   - 做空：D1处于极端下降趋势 → 反弹可能很高 → 不进场
+        threshold = self.d1_trend_strength_max.value / 100.0
+        d1_strength_strong = dataframe[d1_strength_col] > threshold
+
+        long_d1_ok = ~(dataframe[d1_uptrend_col] & d1_strength_strong)   # 极端上升趋势不做多
+        short_d1_ok = ~(dataframe[d1_downtrend_col] & d1_strength_strong) # 极端下降趋势不做空
+
+        # ── v3.2: 区间质量过滤 ──
+        quality_min = self.zone_quality_min.value
+        zone_quality_ok = dataframe["zone_quality"] >= quality_min
+
         # ── 做多：震荡市中，价格靠近支撑位 ──
         long_conds = (
             dataframe["is_ranging"]
             & (dataframe["dist_to_support"] <= entry_zone)
             & (dataframe["dist_to_support"] > 0)
-            & (~dataframe[d1_downtrend_col])
+            & (~dataframe[d1_downtrend_col])        # 原有：D1不能是下降趋势
+            & long_d1_ok                              # v3.2新增：极端上升趋势不做多
+            & zone_quality_ok                         # v3.2新增：区间质量达标
         )
 
-        cooldown = 1  # v3.1: 3→1
+        cooldown = 1
         long_recent = long_conds.rolling(cooldown, min_periods=1).max().shift(1) == 0
         dataframe.loc[long_conds & long_recent, "enter_long"] = 1
 
@@ -292,7 +426,9 @@ class StructureFlowSwingV31(IStrategy):
             dataframe["is_ranging"]
             & (dataframe["dist_to_resistance"] <= entry_zone)
             & (dataframe["dist_to_resistance"] > 0)
-            & (~dataframe[d1_uptrend_col])
+            & (~dataframe[d1_uptrend_col])           # 原有：D1不能是上升趋势
+            & short_d1_ok                             # v3.2新增：极端下降趋势不做空
+            & zone_quality_ok                         # v3.2新增：区间质量达标
         )
 
         short_recent = short_conds.rolling(cooldown, min_periods=1).max().shift(1) == 0
@@ -308,7 +444,7 @@ class StructureFlowSwingV31(IStrategy):
         return dataframe
 
     # ================================================================
-    # 自定义止损：支撑/阻力外侧，ATR*1.5 缓冲
+    # 自定义止损：支撑/阻力外侧，ATR*1.5 缓冲（v3.1逻辑保持不变）
     # ================================================================
 
     def custom_stoploss(
@@ -358,7 +494,7 @@ class StructureFlowSwingV31(IStrategy):
             return min(sl_ratio, 0.20)
 
     # ================================================================
-    # 自定义止盈：区间70%
+    # 自定义止盈：区间70% + v3.2主动退出机制
     # ================================================================
 
     def custom_exit(
@@ -378,6 +514,7 @@ class StructureFlowSwingV31(IStrategy):
 
         last = dataframe.iloc[-1]
 
+        # ── 原有：区间70%止盈 ──
         if not trade.is_short:
             support = last.get("range_support", np.nan)
             resistance = last.get("range_resistance", np.nan)
@@ -397,6 +534,34 @@ class StructureFlowSwingV31(IStrategy):
                 if current_profit >= tp_target:
                     return "take_profit"
 
+        # ── v3.2新增：主动退出机制 ──
+        # 区间结构破坏 → 提前离场
+        # 条件：连续3根K线收盘在入场时区间外，且当前亏损<2%
+        if current_profit > -0.02:
+            # 找到入场时的K线（取最后一根确认的K线，不是当前正在形成的）
+            entry_date = trade.open_date
+            entry_mask = dataframe["date"] <= entry_date
+            if entry_mask.any():
+                entry_idx = dataframe[entry_mask].index[-1]
+                entry_support = dataframe.loc[entry_idx, "range_support"]
+                entry_resistance = dataframe.loc[entry_idx, "range_resistance"]
+
+                if pd.notna(entry_support) and pd.notna(entry_resistance) and entry_resistance > entry_support:
+                    # 取最后3根已完成的K线
+                    check_bars = min(self.early_exit_bars, len(dataframe) - 1)
+                    recent = dataframe.iloc[-(check_bars + 1):-1]  # 排除当前正在形成的K线
+
+                    if len(recent) >= self.early_exit_bars:
+                        outside_count = 0
+                        for _, bar in recent.iterrows():
+                            c = bar["close"]
+                            # 缓冲0.5%避免噪音触发
+                            if c < entry_support * 0.995 or c > entry_resistance * 1.005:
+                                outside_count += 1
+
+                        if outside_count >= self.early_exit_bars:
+                            return "early_exit_structure_broken"
+
         return None
 
     # ================================================================
@@ -414,6 +579,7 @@ class StructureFlowSwingV31(IStrategy):
                 "range": {
                     "is_ranging": {"color": "blue", "type": "line"},
                     "zone_width_pct": {"color": "purple", "type": "line"},
+                    "zone_quality": {"color": "orange", "type": "line"},
                 },
                 "position": {
                     "dist_to_support": {"color": "green", "type": "line"},