253 lines
9.1 KiB
Python
253 lines
9.1 KiB
Python
"""AI 深度分析
|
||
|
||
预先获取 K 线、资金流、技术信号等数据，一次性传入 LLM 生成结构化分析报告。
|
||
不依赖 tool calling，避免 DeepSeek DSML 标签问题。
|
||
"""
|
||
|
||
import asyncio
|
||
import json
|
||
import logging
|
||
import re
|
||
import traceback
|
||
|
||
from app.llm.client import chat_completion
|
||
from app.llm.prompts import TREND_BREAKOUT_ANALYSIS_PROMPT
|
||
from app.config import settings
|
||
|
||
logger = logging.getLogger(__name__)
|
||
|
||
|
||
async def analyze_recommendations(result: dict) -> None:
|
||
"""对所有推荐股票执行 AI 深度分析"""
|
||
recommendations = result.get("recommendations", [])
|
||
if not recommendations or not settings.deepseek_api_key:
|
||
return
|
||
|
||
try:
|
||
await _do_analyze(result, recommendations)
|
||
except Exception as e:
|
||
logger.error(f"AI 分析任务异常: {e}")
|
||
from app.db.error_logger import log_error
|
||
await log_error("analysis_agent", f"AI 分析任务异常: {e}", detail=traceback.format_exc())
|
||
for rec in recommendations:
|
||
if not rec.llm_analysis:
|
||
rec.llm_analysis = "AI 分析暂时不可用"
|
||
await _broadcast_llm_ready(recommendations)
|
||
|
||
|
||
async def _do_analyze(result: dict, recommendations: list) -> None:
|
||
"""分析核心逻辑"""
|
||
market_temp = result.get("market_temp")
|
||
hot_sectors = result.get("hot_sectors", [])
|
||
|
||
# 构建板块文本
|
||
sectors_text = "\n".join(
|
||
f"- {s.sector_name}: 涨幅{s.pct_change}%, 资金流入{s.capital_inflow}万, "
|
||
f"涨停{s.limit_up_count}家, 热度{s.heat_score}分, 阶段={s.stage}"
|
||
for s in hot_sectors[:5]
|
||
) if hot_sectors else "暂无板块数据"
|
||
|
||
# 温度等级
|
||
temp_val = market_temp.temperature if market_temp else 0
|
||
if temp_val >= 60:
|
||
temp_level = "积极"
|
||
elif temp_val >= 30:
|
||
temp_level = "谨慎"
|
||
else:
|
||
temp_level = "低迷"
|
||
|
||
enhanced_count = 0
|
||
for rec in recommendations:
|
||
try:
|
||
# 预先获取该股票的详细数据
|
||
stock_data = await _fetch_stock_data(rec.ts_code, rec.sector)
|
||
|
||
strategy_label = "趋势突破"
|
||
signal_map = {"breakout": "突破型", "pullback": "回踩型", "launch": "启动型", "none": "无信号"}
|
||
entry_label = signal_map.get(rec.entry_signal_type, "无信号")
|
||
system_prompt = TREND_BREAKOUT_ANALYSIS_PROMPT
|
||
|
||
user_msg = _build_user_message(
|
||
rec=rec,
|
||
strategy_label=strategy_label,
|
||
entry_label=entry_label,
|
||
market_temp=market_temp,
|
||
temp_level=temp_level,
|
||
sectors_text=sectors_text,
|
||
stock_data=stock_data,
|
||
)
|
||
|
||
messages = [
|
||
{"role": "system", "content": system_prompt},
|
||
{"role": "user", "content": user_msg},
|
||
]
|
||
|
||
resp = await chat_completion(messages)
|
||
if resp and resp.content:
|
||
analysis = resp.content.strip()
|
||
rec.llm_analysis = analysis
|
||
rec.llm_score = _extract_score(analysis)
|
||
enhanced_count += 1
|
||
else:
|
||
rec.llm_analysis = "AI 分析暂时不可用"
|
||
|
||
except asyncio.CancelledError:
|
||
logger.warning(f"AI 分析 {rec.ts_code} 被取消")
|
||
break
|
||
except Exception as e:
|
||
logger.error(f"AI 分析 {rec.ts_code} 失败: {e}")
|
||
rec.llm_analysis = "AI 分析暂时不可用"
|
||
|
||
# 无论成功失败都保存并广播
|
||
await _save_llm_analysis_to_db(recommendations)
|
||
await _broadcast_llm_ready(recommendations)
|
||
|
||
logger.info(f"AI 深度分析完成: {enhanced_count}/{len(recommendations)} 条")
|
||
|
||
|
||
async def _fetch_stock_data(ts_code: str, sector: str) -> str:
|
||
"""预先获取个股详细数据，拼接为文本供 LLM 分析"""
|
||
from app.llm.tool_executor import (
|
||
_get_stock_kline,
|
||
_get_stock_capital_flow,
|
||
_get_stock_technical_signal,
|
||
_get_sector_performance,
|
||
)
|
||
|
||
parts = []
|
||
|
||
# K 线（最近 30 天摘要）
|
||
try:
|
||
kline_text = await _get_stock_kline(ts_code, 60)
|
||
kline_data = json.loads(kline_text)
|
||
if isinstance(kline_data, list) and kline_data:
|
||
# 只取最近 10 条以控制 token
|
||
recent = kline_data[-10:]
|
||
kline_summary = "\n".join(
|
||
f" {d.get('trade_date', '')}: 收{d.get('close', '')} "
|
||
f"涨跌{d.get('pct_chg', '')}% 量{d.get('vol', '')} "
|
||
f"MA5={d.get('ma5', '')} MA10={d.get('ma10', '')} MA20={d.get('ma20', '')} "
|
||
f"DIF={d.get('dif', '')} DEA={d.get('dea', '')} RSI={d.get('rsi14', '')}"
|
||
for d in recent
|
||
)
|
||
parts.append(f"## K线数据（近10日）\n{kline_summary}")
|
||
except Exception as e:
|
||
logger.debug(f"获取K线数据失败 {ts_code}: {e}")
|
||
|
||
# 资金流向
|
||
try:
|
||
flow_text = await _get_stock_capital_flow(ts_code, 5)
|
||
flow_data = json.loads(flow_text)
|
||
if isinstance(flow_data, list) and flow_data:
|
||
flow_summary = "\n".join(
|
||
f" {d.get('trade_date', '')}: 主力净流入{d.get('main_net_inflow', 0)}万"
|
||
for d in flow_data[-5:]
|
||
)
|
||
parts.append(f"## 资金流向（近5日）\n{flow_summary}")
|
||
except Exception as e:
|
||
logger.debug(f"获取资金流向失败 {ts_code}: {e}")
|
||
|
||
# 技术信号
|
||
try:
|
||
signal_text = await _get_stock_technical_signal(ts_code)
|
||
parts.append(f"## 技术信号\n{signal_text}")
|
||
except Exception as e:
|
||
logger.debug(f"获取技术信号失败 {ts_code}: {e}")
|
||
|
||
# 板块表现
|
||
if sector:
|
||
try:
|
||
sector_text = await _get_sector_performance(sector)
|
||
parts.append(f"## 板块数据\n{sector_text}")
|
||
except Exception as e:
|
||
logger.debug(f"获取板块数据失败 {sector}: {e}")
|
||
|
||
return "\n\n".join(parts) if parts else "暂无额外数据"
|
||
|
||
|
||
def _build_user_message(
|
||
rec,
|
||
strategy_label: str,
|
||
entry_label: str,
|
||
market_temp,
|
||
temp_level: str,
|
||
sectors_text: str,
|
||
stock_data: str,
|
||
) -> str:
|
||
"""构建完整的用户消息（含预获取的数据）"""
|
||
return f"""## 量化系统数据
|
||
- 股票: {rec.name}({rec.ts_code})
|
||
- 所属板块: {rec.sector}
|
||
- 策略类型: {strategy_label}
|
||
- 入场信号: {entry_label}
|
||
- 综合评分: {rec.score}分（{rec.level}）
|
||
- 各维度: 市场{rec.market_temp_score} | 板块{rec.sector_score} | 资金{rec.capital_score} | 技术{rec.technical_score} | 位置{rec.position_score} | 估值{rec.valuation_score}
|
||
- 信号: {rec.signal}
|
||
- 参考价: 入场{rec.entry_price or 'N/A'} / 目标{rec.target_price or 'N/A'} / 止损{rec.stop_loss or 'N/A'}
|
||
- 量化理由: {"；".join(rec.reasons) if rec.reasons else "无"}
|
||
|
||
## 市场环境
|
||
- 市场温度: {market_temp.temperature if market_temp else 'N/A'}/100（{temp_level}）
|
||
- 涨跌比: {market_temp.up_count if market_temp else 0}涨 / {market_temp.down_count if market_temp else 0}跌
|
||
- 涨停: {market_temp.limit_up_count if market_temp else 0}家
|
||
|
||
## 热门板块
|
||
{sectors_text}
|
||
|
||
## 个股详细数据
|
||
{stock_data}
|
||
|
||
请根据以上所有数据，按照指定格式输出深度分析报告。"""
|
||
|
||
|
||
def _extract_score(text: str) -> float | None:
|
||
"""从 AI 分析报告中提取评分（1-10）"""
|
||
match = re.search(r"###\s*AI\s*评分[^\d]*(\d+(?:\.\d+)?)", text)
|
||
if match:
|
||
score = float(match.group(1))
|
||
return min(max(score, 1), 10)
|
||
return None
|
||
|
||
|
||
async def _save_llm_analysis_to_db(recommendations: list) -> None:
|
||
"""将 AI 分析结果更新到数据库"""
|
||
try:
|
||
from app.db.database import get_db
|
||
from sqlalchemy import text
|
||
|
||
async with get_db() as db:
|
||
for rec in recommendations:
|
||
if not rec.llm_analysis:
|
||
continue
|
||
await db.execute(
|
||
text(
|
||
"UPDATE recommendations SET llm_analysis = :analysis, "
|
||
"llm_score = :score "
|
||
"WHERE ts_code = :code AND date(created_at) = date('now', 'localtime') "
|
||
"AND scan_session = :session"
|
||
),
|
||
{
|
||
"analysis": rec.llm_analysis,
|
||
"score": rec.llm_score,
|
||
"code": rec.ts_code,
|
||
"session": rec.scan_session,
|
||
},
|
||
)
|
||
await db.commit()
|
||
except Exception as e:
|
||
logger.error(f"保存 AI 分析到数据库失败: {e}")
|
||
from app.db.error_logger import log_error
|
||
await log_error("analysis_agent", f"保存 AI 分析到数据库失败: {e}", detail=traceback.format_exc())
|
||
|
||
|
||
async def _broadcast_llm_ready(recommendations: list) -> None:
|
||
"""通过 WebSocket 广播 AI 分析完成事件"""
|
||
try:
|
||
from app.api.websocket import broadcast_update
|
||
await broadcast_update({
|
||
"type": "llm_analysis_ready",
|
||
"count": len([r for r in recommendations if r.llm_analysis]),
|
||
})
|
||
except Exception as e:
|
||
logger.error(f"广播 AI 分析完成失败: {e}")
|