import numpy as np import talib from scipy.signal import stft from datetime import datetime, timedelta from typing import Optional, Any, List, Dict from src.core_data import Bar, Order from src.indicators.base_indicators import Indicator from src.indicators.indicators import Empty from src.strategies.base_strategy import Strategy class SpectralTrendStrategy(Strategy): """ 频域能量相变策略 - 极简回归版 核心哲学: 1. 频域 (STFT): 负责"判势" —— 现在的市场是震荡(噪音主导)还是趋势(低频主导)？ 2. 时域 (Regression): 负责"定向" —— 这个低频趋势是向上的还是向下的？ 这种组合避免了频域相位计算的复杂性和不稳定性，回归了量化的本质。 """ def __init__( self, context: Any, main_symbol: str, enable_log: bool, trade_volume: int, # --- 市场参数 --- bars_per_day: int = 23, # --- 策略参数 --- spectral_window_days: float = 2.0, low_freq_days: float = 2.0, high_freq_days: float = 1.0, trend_strength_threshold: float = 0.2, # 强度阈值 exit_threshold: float = 0.1, # 退出阈值 slope_threshold: float = 0.0, # 斜率阈值 (0.05表示每根K线移动0.05个标准差) max_hold_days: int = 10, # --- 其他 --- order_direction: Optional[List[str]] = None, indicators: Indicator = None, model_indicator: Indicator = None, reverse: bool = False, ): super().__init__(context, main_symbol, enable_log) if order_direction is None: order_direction = ['BUY', 'SELL'] self.trade_volume = trade_volume self.bars_per_day = bars_per_day self.spectral_window_days = spectral_window_days self.low_freq_days = low_freq_days self.high_freq_days = high_freq_days self.trend_strength_threshold = trend_strength_threshold self.exit_threshold = exit_threshold self.slope_threshold = slope_threshold self.max_hold_days = max_hold_days self.order_direction = order_direction self.model_indicator = model_indicator or Empty() self.indicators = indicators or Empty() self.reverse = reverse # 计算窗口大小 self.spectral_window = int(self.spectral_window_days * self.bars_per_day) # 确保偶数 (STFT偏好) if self.spectral_window % 2 != 0: self.spectral_window += 1 # 频率边界 self.low_freq_bound = 1.0 / self.low_freq_days if self.low_freq_days > 0 else float('inf') self.high_freq_bound = 1.0 / self.high_freq_days if self.high_freq_days > 0 else 0.0 self.order_id_counter = 0 self.entry_time = None self.position_direction = None self.log(f"SpectralTrendStrategy (Regression) Init. Window: {self.spectral_window} bars") def on_open_bar(self, open_price: float, symbol: str): self.symbol = symbol bar_history = self.get_bar_history() current_time = self.get_current_time() self.cancel_all_pending_orders(self.main_symbol) if len(bar_history) < self.spectral_window + 5: return # 强制平仓检查 if self.entry_time and (current_time - self.entry_time) >= timedelta(days=self.max_hold_days): self.close_all_positions() self.entry_time = None self.position_direction = None return # 获取数据并归一化 closes = np.array([b.close for b in bar_history[-self.spectral_window:]], dtype=float) # 计算核心指标 trend_strength, trend_slope = self.calculate_market_state(closes) position_volume = self.get_current_positions().get(self.symbol, 0) if self.trading: if position_volume == 0: self.evaluate_entry_signal(open_price, trend_strength, trend_slope) else: self.manage_open_position(position_volume, trend_strength, trend_slope) def calculate_market_state(self, prices: np.array) -> (float, float): """ 【显式傅里叶】计算低频能量占比 (完全参数化) 步骤: 1. 价格归一化 (窗口内) 2. 短时傅里叶变换 (STFT) - 采样率=bars_per_day 3. 动态计算频段边界 (基于bars_per_day) 4. 趋势强度 = 低频能量 / (低频+高频能量) """ # 1. 验证数据长度 if len(prices) < self.spectral_window: return 0.0, 0.0 # 2. 价格归一化 (仅使用窗口内数据) window_data = prices[-self.spectral_window:] normalized = (window_data - np.mean(window_data)) / (np.std(window_data) + 1e-8) normalized = normalized[-self.spectral_window:] # 3. STFT (采样率=bars_per_day) try: # fs: 每天的样本数 (bars_per_day) f, t, Zxx = stft( normalized, fs=self.bars_per_day, # 关键: 适配市场结构 nperseg=self.spectral_window, noverlap=max(0, self.spectral_window // 2), boundary=None, padded=False ) except Exception as e: self.log(f"STFT calculation error: {str(e)}") return 0.0, 0.0 # 4. 过滤无效频率 (STFT返回频率范围: 0 到 fs/2) valid_mask = (f >= 0) & (f <= self.bars_per_day / 2) f = f[valid_mask] Zxx = Zxx[valid_mask, :] if Zxx.size == 0 or Zxx.shape[1] == 0: return 0.0, 0.0 # 5. 计算最新时间点的能量 current_energy = np.abs(Zxx[:, -1]) ** 2 # 6. 动态频段定义 (cycles/day) # 低频: 周期 > low_freq_days → 频率 < 1/low_freq_days low_freq_mask = f < self.low_freq_bound # 高频: 周期 < high_freq_days → 频率 > 1/high_freq_days high_freq_mask = f > self.high_freq_bound # 7. 能量计算 low_energy = np.sum(current_energy[low_freq_mask]) if np.any(low_freq_mask) else 0.0 high_energy = np.sum(current_energy[high_freq_mask]) if np.any(high_freq_mask) else 0.0 total_energy = low_energy + high_energy + 1e-8 # 防除零 # 8. 趋势强度 = 低频能量占比 trend_strength = low_energy / total_energy # --- 3. 时域分析 (Regression) - 只负责"方向" --- # 使用最小二乘法拟合一条直线 y = kx + b # x 是时间序列 [0, 1, 2...], y 是归一化价格 # slope 代表：每经过一根K线，价格变化多少个标准差 x = np.arange(len(normalized)) slope, intercept = np.polyfit(x, normalized, 1) return trend_strength, slope def evaluate_entry_signal(self, open_price: float, trend_strength: float, trend_slope: float): """ 入场逻辑： 当频域告诉我们"有趋势"(Strength高)，且时域告诉我们"方向明确"(Slope陡峭)时入场。 """ # 1. 滤除噪音震荡 (STFT关卡) if trend_strength > self.trend_strength_threshold: direction = None # 2. 确认方向 (回归关卡) # slope > 0.05 意味着趋势向上且有一定力度 if "BUY" in self.order_direction and trend_slope > self.slope_threshold: direction = "BUY" # slope < -0.05 意味着趋势向下且有一定力度 elif "SELL" in self.order_direction and trend_slope < -self.slope_threshold: direction = "SELL" if direction: # 辅助指标过滤 if not self.indicators.is_condition_met(*self.get_indicator_tuple()): return # 反向逻辑 direction = direction if not self.model_indicator.is_condition_met(*self.get_indicator_tuple()): direction = "SELL" if direction == "BUY" else "BUY" if self.reverse: direction = "SELL" if direction == "BUY" else "BUY" self.log(f"Signal: {direction} | Strength={trend_strength:.2f} | Slope={trend_slope:.4f}") self.send_limit_order(direction, open_price, self.trade_volume, "OPEN") self.entry_time = self.get_current_time() self.position_direction = "LONG" if direction == "BUY" else "SHORT" def manage_open_position(self, volume: int, trend_strength: float, trend_slope: float): """ 离场逻辑： 仅依赖频域能量。只要低频能量依然主导，说明趋势（无论方向）未被破坏。 一旦能量降到 exit_threshold 以下，说明市场进入混乱/震荡，离场观望。 """ if trend_strength < self.exit_threshold: direction = "CLOSE_LONG" if volume > 0 else "CLOSE_SHORT" self.log(f"Exit: {direction} | Strength={trend_strength:.2f} < {self.exit_threshold}") self.close_position(direction, abs(volume)) self.entry_time = None self.position_direction = None # --- 交易辅助 --- def close_all_positions(self): positions = self.get_current_positions() if self.symbol in positions and positions[self.symbol] != 0: dir = "CLOSE_LONG" if positions[self.symbol] > 0 else "CLOSE_SHORT" self.close_position(dir, abs(positions[self.symbol])) def close_position(self, direction: str, volume: int): self.send_market_order(direction, volume, offset="CLOSE") def send_market_order(self, direction: str, volume: int, offset: str): order_id = f"{self.symbol}_{direction}_MKT_{self.order_id_counter}" self.order_id_counter += 1 order = Order( id=order_id, symbol=self.symbol, direction=direction, volume=volume, price_type="MARKET", submitted_time=self.get_current_time(), offset=offset ) self.send_order(order) def send_limit_order(self, direction: str, limit_price: float, volume: int, offset: str): order_id = f"{self.symbol}_{direction}_LMT_{self.order_id_counter}" self.order_id_counter += 1 order = Order( id=order_id, symbol=self.symbol, direction=direction, volume=volume, price_type="LIMIT", submitted_time=self.get_current_time(), offset=offset, limit_price=limit_price ) self.send_order(order)