""" 自动化回测服务 - 任务执行器 调用 vnpy 原生 BacktestingEngine 执行回测 """ import os import sys import tempfile import traceback from datetime import datetime from typing import Optional import matplotlib.pyplot as plt import pandas as pd from vnpy.trader.engine import MainEngine from vnpy.trader.event import EventEngine from vnpy.trader.backtesting import BacktestingEngine from vnpy.trader.constant import Interval from vnpy.trader.database import database_manager from vnpy.trader.object import HistoryRequest from .config import settings from .models import BacktestTask, BacktestResult, BacktestStatistics, TaskStatus, BacktestTaskWithId from .result_storage import storage INTERVAL_MAP = { "1m": Interval.MINUTE, "5m": Interval.FIVE_MINUTE, "15m": Interval.FIFTEEN_MINUTE, "30m": Interval.THIRTY_MINUTE, "1h": Interval.HOUR, "4h": Interval.FOUR_HOUR, "1d": Interval.DAILY, "1w": Interval.WEEKLY, } class BacktestExecutor: """回测任务执行器""" def __init__(self): pass def _load_strategy(self, task: BacktestTask): """动态加载策略代码""" # 将策略代码写入临时文件 strategy_code = task.strategy_code # 创建临时目录 temp_dir = tempfile.mkdtemp() sys.path.insert(0, temp_dir) strategy_file = os.path.join(temp_dir, "strategy.py") with open(strategy_file, "w", encoding="utf-8") as f: f.write(strategy_code) # 导入模块 import importlib spec = importlib.util.spec_from_file_location("dynamic_strategy", strategy_file) module = importlib.util.module_from_spec(spec) sys.modules["dynamic_strategy"] = module spec.loader.exec_module(module) # 找到策略类 - 假设第一个继承自 Strategy 的就是我们要的 from vnpy.trader.strategy import Strategy strategy_class = None for attr_name in dir(module): attr = getattr(module, attr_name) if isinstance(attr, type) and issubclass(attr, Strategy) and attr != Strategy: strategy_class = attr break if not strategy_class: raise ValueError("策略代码中没有找到 Strategy 子类，请检查策略代码") # 创建策略实例，注入参数 strategy = strategy_class return strategy def execute_backtest(self, task: BacktestTaskWithId) -> BacktestResult: """执行一次回测""" from vnpy.trader.database import database_manager start_time = datetime.now() started_at = start_time.isoformat() # 更新任务状态为运行中 task.status = TaskStatus.RUNNING task.started_at = started_at storage.save_task(task) result = BacktestResult( task_id=task.task_id, strategy_name=task.strategy_name, status=TaskStatus.RUNNING, result_csv_path="", created_at=task.created_at, started_at=started_at, ) try: # 加载策略类 strategy_class = self._load_strategy(task) # 获取interval interval = INTERVAL_MAP.get(task.interval, Interval.DAILY) # 查询历史数据 - 使用 vnpy 数据库 req = HistoryRequest( symbol=task.symbol, exchange=None, # 由代码处理 interval=interval, start=task.start_date, end=task.end_date, ) data = database_manager.query_history(req) if data.empty: raise ValueError(f"未找到 {task.symbol} 在 [{task.start_date}, {task.end_date}] 范围内的历史数据") # 创建回测引擎 engine = BacktestingEngine() # 设置参数 engine.set_parameters( data=data, interval=interval, capital=task.capital, tick_size=task.tick_size, ) # 添加策略 engine.add_strategy(strategy_class, task.parameters) # 运行回测 engine.run_backtesting() # 计算统计结果 df = engine.calculate_results() # 统计结果 statistics = engine.get_result_statistics() # 转换为我们的数据模型 stats = BacktestStatistics( start_date=statistics["start_date"].isoformat() if hasattr(statistics["start_date"], "isoformat") else str(statistics["start_date"]), end_date=statistics["end_date"].isoformat() if hasattr(statistics["end_date"], "isoformat") else str(statistics["end_date"]), total_days=int(statistics["total_days"]), total_trades=int(statistics["total_trades"]), winning_trades=int(statistics["winning_trades"]), losing_trades=int(statistics["losing_trades"]), win_rate=float(statistics["win_rate"]), total_return=float(statistics["total_return"]), annual_return=float(statistics["annual_return"]), sharpe_ratio=float(statistics["sharpe_ratio"]), max_drawdown=float(statistics["max_drawdown"]), profit_factor=float(statistics.get("profit_factor", 0)), calmar_ratio=float(statistics.get("calmar_ratio", 0)), ) # 保存净值CSV result_csv_path = storage.get_task_path(task.task_id, TaskStatus.COMPLETED, "equity.csv") os.makedirs(os.path.dirname(result_csv_path), exist_ok=True) df.to_csv(result_csv_path, index=False) # 绘制收益曲线 png_path = storage.get_task_path(task.task_id, TaskStatus.COMPLETED, "equity_curve.png") self._plot_equity_curve(df, png_path) # 保存成交记录 trades = engine.get_trades() if trades: trades_df = pd.DataFrame([t.__dict__ for t in trades]) trades_csv_path = storage.get_task_path(task.task_id, TaskStatus.COMPLETED, "trades.csv") trades_df.to_csv(trades_csv_path, index=False) else: trades_csv_path = None # 完成结果 result.status = TaskStatus.COMPLETED result.statistics = stats result.result_csv_path = result_csv_path result.equity_curve_png_path = png_path result.trades_csv_path = trades_csv_path completed_at = datetime.now().isoformat() result.completed_at = completed_at storage.save_result(result) return result except Exception as e: # 捕获异常，记录错误信息 error_msg = f"{str(e)}\n{traceback.format_exc()}" result.status = TaskStatus.FAILED result.error_message = error_msg completed_at = datetime.now().isoformat() result.completed_at = completed_at storage.save_result(result) return result def _plot_equity_curve(self, df: pd.DataFrame, output_path: str): """绘制收益曲线""" plt.figure(figsize=(12, 6)) if "equity" in df.columns: plt.plot(df.index, df["equity"], label="净值曲线", linewidth=2) elif "net_pnl" in df.columns: cumulative = df["net_pnl"].cumsum() plt.plot(df.index, cumulative, label="累计收益", linewidth=2) plt.title("回测收益曲线") plt.xlabel("时间") plt.ylabel("净值") plt.grid(True, alpha=0.3) plt.legend() plt.tight_layout() plt.savefig(output_path, dpi=150) plt.close() executor = BacktestExecutor()