feat(strategy): Implement signal filtering mechanism for trade signals
Added a unified signal filtering layer to the trading strategy, allowing for enhanced analysis of trade signals based on market conditions. The filtering process includes logging reasons for acceptance or rejection, improving transparency and decision-making in automated trading. Exception handling is also implemented to ensure robustness during signal filtering application.
This commit is contained in:
薇薇安
parent
199c4a95dd
commit
d68d3ad66c
215
trading_system/signal_filters.py
Normal file
215
trading_system/signal_filters.py
Normal file
|
|
@ -0,0 +1,215 @@
|
|||
"""
|
||||
信号过滤插件层
|
||||
|
||||
目标：
|
||||
- 将零散的信号过滤规则抽象为可插拔的「过滤器」，便于后续演进（新规则 / 机器学习模型）。
|
||||
- 第一阶段：在不大幅改动既有策略的前提下，增加一层统一的信号过滤入口。
|
||||
|
||||
设计原则：
|
||||
- 只做「进一步过滤/收紧」，不绕过任何已有风控与止损规则。
|
||||
- 过滤失败时只会减少开仓次数，不会放宽任何限制。
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
from dataclasses import dataclass
|
||||
from typing import Dict, List, Optional
|
||||
|
||||
from . import config
|
||||
|
||||
|
||||
@dataclass
|
||||
class SignalFilterContext:
|
||||
"""
|
||||
信号过滤上下文
|
||||
|
||||
说明：
|
||||
- symbol_info 与 trade_signal 来自现有 strategy / scanner 逻辑，不做修改。
|
||||
- 仅在需要时读取 config.TRADING_CONFIG，避免在过滤层重复维护配置。
|
||||
"""
|
||||
|
||||
symbol: str
|
||||
symbol_info: Dict
|
||||
trade_signal: Dict
|
||||
market_regime: Optional[str] = None
|
||||
|
||||
|
||||
@dataclass
|
||||
class SignalFilterResult:
|
||||
"""信号过滤结果"""
|
||||
|
||||
accepted: bool
|
||||
reasons: List[str]
|
||||
|
||||
|
||||
async def apply_signal_filters(ctx: SignalFilterContext) -> SignalFilterResult:
|
||||
"""
|
||||
应用所有已启用的信号过滤规则。
|
||||
|
||||
第一阶段：
|
||||
- 引入「最小波动率」过滤：在波动极低（接近死币）时不自动开仓，仅保留推荐。
|
||||
- 其它过滤（资金费率、流动性、RSI 等）仍由现有 strategy 逻辑负责。
|
||||
|
||||
后续阶段预留：
|
||||
- 可在此处串联更多插件式过滤器（如多周期趋势评分、ML 评分过滤等）。
|
||||
"""
|
||||
reasons: List[str] = []
|
||||
accepted = True
|
||||
|
||||
# 1) 波动率过滤：避免在接近「死币」的环境下做趋势单
|
||||
if not _filter_by_volatility(ctx, reasons):
|
||||
accepted = False
|
||||
|
||||
# 2) 严格成交量过滤：在自动交易阶段只做流动性更好的标的
|
||||
if accepted and not _filter_by_strict_volume(ctx, reasons):
|
||||
accepted = False
|
||||
|
||||
# 3) 按市场状态提高最低信号强度：ranging/unknown 环境下只做更强信号
|
||||
if accepted and not _filter_by_regime_and_signal_strength(ctx, reasons):
|
||||
accepted = False
|
||||
|
||||
# TODO(后续阶段)：在此处串联更多过滤器，例如：
|
||||
# - _filter_by_multi_timeframe_trend(ctx, reasons)
|
||||
# - await _filter_by_ml_score(ctx, reasons)
|
||||
|
||||
return SignalFilterResult(accepted=accepted, reasons=reasons)
|
||||
|
||||
|
||||
def _filter_by_volatility(ctx: SignalFilterContext, reasons: List[str]) -> bool:
|
||||
"""
|
||||
基于主周期涨跌幅的简单波动率过滤。
|
||||
|
||||
实现细节：
|
||||
- 使用 scanner 产出的 kline_change_percent（主周期涨跌幅）作为代理波动率。
|
||||
- MIN_VOLATILITY 采用比例表示：0.03 = 3%。
|
||||
- 仅当有明确 should_trade 信号时才应用此过滤，避免影响推荐展示。
|
||||
"""
|
||||
if not ctx.trade_signal.get("should_trade"):
|
||||
return True
|
||||
|
||||
try:
|
||||
min_volatility = float(config.TRADING_CONFIG.get("MIN_VOLATILITY", 0.0) or 0.0)
|
||||
except Exception:
|
||||
min_volatility = 0.0
|
||||
|
||||
if min_volatility <= 0:
|
||||
return True
|
||||
|
||||
change_pct = ctx.symbol_info.get("kline_change_percent")
|
||||
if change_pct is None:
|
||||
# 无主周期涨跌数据时不强制过滤，仅记录调试信息可在后续完善
|
||||
return True
|
||||
|
||||
try:
|
||||
change_pct_val = float(change_pct)
|
||||
except (TypeError, ValueError):
|
||||
return True
|
||||
|
||||
# kline_change_percent 为百分数（如 3.5 表示 +3.5%）
|
||||
realized_vol = abs(change_pct_val) / 100.0
|
||||
if realized_vol < min_volatility:
|
||||
reasons.append(
|
||||
f"❌ 波动率过滤：主周期波动率≈{realized_vol:.3%} < 最小要求 {min_volatility:.3%}"
|
||||
)
|
||||
return False
|
||||
|
||||
return True
|
||||
|
||||
|
||||
def _filter_by_strict_volume(ctx: SignalFilterContext, reasons: List[str]) -> bool:
|
||||
"""
|
||||
严格成交量过滤：在自动交易阶段只做 24h 成交额更高的一批标的。
|
||||
|
||||
说明：
|
||||
- 使用 MIN_VOLUME_24H_STRICT（默认 5000 万 USDT）作为更严格的流动性门槛。
|
||||
- 仅当 should_trade=True 时生效，不影响推荐与扫描排序。
|
||||
"""
|
||||
if not ctx.trade_signal.get("should_trade"):
|
||||
return True
|
||||
|
||||
volume_24h = ctx.symbol_info.get("volume24h")
|
||||
if volume_24h is None:
|
||||
return True
|
||||
|
||||
try:
|
||||
volume_val = float(volume_24h)
|
||||
except (TypeError, ValueError):
|
||||
return True
|
||||
|
||||
try:
|
||||
strict_min = float(config.TRADING_CONFIG.get("MIN_VOLUME_24H_STRICT", 0) or 0)
|
||||
except Exception:
|
||||
strict_min = 0.0
|
||||
|
||||
if strict_min <= 0:
|
||||
return True
|
||||
|
||||
if volume_val < strict_min:
|
||||
reasons.append(
|
||||
f"❌ 严格成交量过滤：24h成交额≈{volume_val:,.0f} < 严格下限 {strict_min:,.0f}"
|
||||
)
|
||||
return False
|
||||
|
||||
return True
|
||||
|
||||
|
||||
def _filter_by_regime_and_signal_strength(ctx: SignalFilterContext, reasons: List[str]) -> bool:
|
||||
"""
|
||||
按市场状态提高最低信号强度。
|
||||
|
||||
设计：
|
||||
- 在 trending 状态下沿用原有 MIN_SIGNAL_STRENGTH 逻辑。
|
||||
- 在 ranging/unknown 状态下，即使通过基础门槛，也要求信号更强才放行。
|
||||
这样可以减少边缘信号在“脏行情”里的出手次数。
|
||||
"""
|
||||
if not ctx.trade_signal.get("should_trade"):
|
||||
return True
|
||||
|
||||
direction = ctx.trade_signal.get("direction")
|
||||
if direction not in ("BUY", "SELL"):
|
||||
return True
|
||||
|
||||
try:
|
||||
strength = int(ctx.trade_signal.get("strength") or 0)
|
||||
except (TypeError, ValueError):
|
||||
strength = 0
|
||||
|
||||
mr = (ctx.market_regime or "").strip().lower() if ctx.market_regime else ""
|
||||
|
||||
# 默认仅对 ranging/unknown 提升门槛；门槛值可配置
|
||||
try:
|
||||
extra_ranging = int(config.TRADING_CONFIG.get("RANGING_EXTRA_MIN_STRENGTH", 0) or 0)
|
||||
extra_unknown = int(config.TRADING_CONFIG.get("UNKNOWN_EXTRA_MIN_STRENGTH", 0) or 0)
|
||||
base_min = int(config.TRADING_CONFIG.get("MIN_SIGNAL_STRENGTH", 7) or 7)
|
||||
except Exception:
|
||||
extra_ranging = 0
|
||||
extra_unknown = 0
|
||||
base_min = 7
|
||||
|
||||
target_min = base_min
|
||||
bucket = None
|
||||
if mr == "ranging" and extra_ranging > 0:
|
||||
target_min = max(target_min, base_min + extra_ranging)
|
||||
bucket = "震荡市"
|
||||
elif mr not in ("trending", "ranging") and extra_unknown > 0:
|
||||
target_min = max(target_min, base_min + extra_unknown)
|
||||
bucket = "未知市况"
|
||||
|
||||
if bucket and strength < target_min:
|
||||
reasons.append(
|
||||
f"❌ {bucket}信号强度过滤：强度={strength}/10 < 有效门槛 {target_min}/10"
|
||||
)
|
||||
return False
|
||||
|
||||
return True
|
||||
|
||||
|
||||
async def apply_ml_score_filter_placeholder(ctx: SignalFilterContext) -> SignalFilterResult:
|
||||
"""
|
||||
预留的 ML 评分过滤入口（占位实现）。
|
||||
|
||||
说明：
|
||||
- 当前实现总是返回接受，且不修改交易行为。
|
||||
- 后续可在此处加载离线训练好的模型，对特征向量打分并决定是否放行。
|
||||
"""
|
||||
return SignalFilterResult(accepted=True, reasons=[])
|
||||
|
||||
|
|
@ -11,6 +11,7 @@ try:
|
|||
from .market_scanner import MarketScanner
|
||||
from .risk_manager import RiskManager
|
||||
from .position_manager import PositionManager
|
||||
from .signal_filters import SignalFilterContext, apply_signal_filters
|
||||
from . import config
|
||||
from .symbol_policy import resolve_symbol_trading_policy
|
||||
except ImportError:
|
||||
|
|
@ -18,6 +19,7 @@ except ImportError:
|
|||
from market_scanner import MarketScanner
|
||||
from risk_manager import RiskManager
|
||||
from position_manager import PositionManager
|
||||
from signal_filters import SignalFilterContext, apply_signal_filters
|
||||
import config
|
||||
from symbol_policy import resolve_symbol_trading_policy
|
||||
|
||||
|
|
@ -213,7 +215,32 @@ class TradingStrategy:
|
|||
f"{symbol} 技术指标分析: {trade_signal['reason']}, 跳过自动交易"
|
||||
)
|
||||
continue
|
||||
|
||||
|
||||
# 统一的信号过滤插件层（第一阶段：波动率等「进一步收紧」过滤）
|
||||
try:
|
||||
sf_ctx = SignalFilterContext(
|
||||
symbol=symbol,
|
||||
symbol_info=symbol_info,
|
||||
trade_signal=trade_signal,
|
||||
market_regime=market_regime,
|
||||
)
|
||||
sf_result = await apply_signal_filters(sf_ctx)
|
||||
if not sf_result.accepted:
|
||||
# 将过滤原因拼接到 trade_signal 的 reason 里，便于后续分析
|
||||
extra_reason = " | ".join(sf_result.reasons) if sf_result.reasons else "信号过滤未通过"
|
||||
base_reason = trade_signal.get("reason") or ""
|
||||
combined_reason = f"{base_reason} | {extra_reason}" if base_reason else extra_reason
|
||||
trade_signal["reason"] = combined_reason
|
||||
logger.info(f\"{symbol} 信号过滤未通过，跳过自动交易: {combined_reason}\")
|
||||
continue
|
||||
elif sf_result.reasons:
|
||||
# 若过滤层仅追加了评语（accepted=True），也拼接到 reason 便于排查
|
||||
extra_reason = " | ".join(sf_result.reasons)
|
||||
base_reason = trade_signal.get("reason") or ""
|
||||
trade_signal["reason"] = f"{base_reason} | {extra_reason}" if base_reason else extra_reason
|
||||
except Exception as e:
|
||||
logger.warning(f\"{symbol} 应用信号过滤插件时出错（忽略，回退为原逻辑）: {e}\")
|
||||
|
||||
# 确定交易方向（基于技术指标）
|
||||
trade_direction = trade_signal['direction']
|
||||
# 4H 下跌禁止开多（熊市/保守方案：避免逆势抄底）
|
||||
|
|
|
|||
Loading…
Reference in New Issue
Block a user