feat: AI事前预警 — 现金流预测+预警扩展+准确率+情景建议

backend/app/utils/cash_forecast_engine.py

@@ -0,0 +1,322 @@
+"""现金流预测引擎 — 根据历史KPI数据推算未来30天现金流"""
+import logging
+from datetime import datetime, timedelta
+from typing import Optional
+from sqlalchemy.orm import Session
+import math
+import random
+
+logger = logging.getLogger("cma.cash_forecast")
+
+# 默认现金阈值（万元）
+DEFAULT_CASH_WARNING = 20.0   # 黄灯 — 低于20万
+DEFAULT_CASH_CRITICAL = 10.0  # 红灯 — 低于10万
+
+# 历史KPI编码映射
+KPI_CODES = {
+    "operating_cash_flow": "CASH_FLOW_001",   # 经营现金流
+    "receivables": "AR_001",                    # 应收账款
+    "payables": "AP_001",                      # 应付账款
+    "cash_balance": "CASH_001",                # 现金余额
+}
+
+
+def get_entity_kpi_history(entity_id: int, kpi_code: str, db: Session, limit_months: int = 6) -> list:
+    """获取实体某个KPI的历史值"""
+    from app.models import KPIDefinition, KPIValue
+    kpi = db.query(KPIDefinition).filter(
+        KPIDefinition.kpi_code == kpi_code,
+        KPIDefinition.entity_id == entity_id,
+    ).first()
+    if not kpi:
+        return []
+    values = db.query(KPIValue).filter(
+        KPIValue.kpi_id == kpi.id,
+        KPIValue.actual_value.isnot(None),
+    ).order_by(KPIValue.period.desc()).limit(limit_months).all()
+    return values
+
+
+def calc_trend(values: list) -> float:
+    """计算趋势系数 — 线性回归斜率 / 均值"""
+    if len(values) < 2:
+        return 0.0
+    vals = [v.actual_value for v in values]
+    n = len(vals)
+    avg_x = (n - 1) / 2.0
+    avg_y = sum(vals) / n
+    num = sum((i - avg_x) * (vals[i] - avg_y) for i in range(n))
+    den = sum((i - avg_x) ** 2 for i in range(n))
+    slope = num / den if den != 0 else 0
+    return slope / max(abs(avg_y), 1.0) * 100  # 趋势百分比
+
+
+def forecast_cash_flow(
+    entity_id: int,
+    db: Session,
+    days: int = 30,
+    current_cash: Optional[float] = None,
+) -> dict:
+    """
+    预测未来30天现金流
+    
+    算法：
+    1. 获取历史经营现金流、应收、应付趋势
+    2. 推算每日现金流入/流出
+    3. 生成每日预测值+置信区间
+    """
+    from app.models import KPIDefinition, KPIValue
+
+    # 获取当前现金余额
+    if current_cash is None:
+        cash_kpi = db.query(KPIDefinition).filter(
+            KPIDefinition.kpi_code == KPI_CODES["cash_balance"],
+            KPIDefinition.entity_id == entity_id,
+        ).first()
+        if cash_kpi:
+            latest_cash = db.query(KPIValue).filter(
+                KPIValue.kpi_id == cash_kpi.id,
+                KPIValue.actual_value.isnot(None),
+            ).order_by(KPIValue.period.desc()).first()
+            base_cash = latest_cash.actual_value if latest_cash else 30.0
+        else:
+            base_cash = 30.0  # 默认假设30万
+    else:
+        base_cash = current_cash
+
+    # 获取经营现金流历史
+    ocf_history = get_entity_kpi_history(entity_id, KPI_CODES["operating_cash_flow"], db)
+    ocf_trend = calc_trend(ocf_history)
+
+    # 获取应收历史
+    ar_history = get_entity_kpi_history(entity_id, KPI_CODES["receivables"], db)
+    ar_trend = calc_trend(ar_history)
+
+    # 获取应付历史
+    ap_history = get_entity_kpi_history(entity_id, KPI_CODES["payables"], db)
+    ap_trend = calc_trend(ap_history)
+
+    # 计算日均现金变化
+    ocf_avg = sum(v.actual_value for v in ocf_history) / max(len(ocf_history), 1) / 30.0 if ocf_history else 0.5
+
+    # 预测逻辑：趋势影响 + 季节性（月底回款高峰）
+    forecast = []
+    cash = base_cash
+    today = datetime.now().replace(hour=0, minute=0, second=0, microsecond=0)
+
+    for day_offset in range(1, days + 1):
+        d = today + timedelta(days=day_offset)
+        day_of_month = d.day
+        is_month_end = day_of_month >= 25
+
+        # 每日现金变化 = 经营现金流日均值 × (1 + 趋势调整) + 季节因子
+        trend_factor = 1.0 + ocf_trend / 100.0
+        daily_change = ocf_avg * trend_factor
+
+        # 月底回款高峰
+        if is_month_end:
+            daily_change += ocf_avg * 0.3  # 月底多30%回款
+
+        # 周末效应
+        if d.weekday() >= 5:
+            daily_change *= 0.5  # 周末收支减半
+
+        cash += daily_change
+
+        # 置信区间：随时间增加而扩大
+        confidence_band = 1.0 + day_offset * 0.08  # 每过1天，区间扩大8%
+        std = max(abs(daily_change) * confidence_band, 0.5)
+        lower = cash - std * 0.5
+        upper = cash + std * 0.5
+
+        # 预警状态
+        if cash < DEFAULT_CASH_CRITICAL:
+            status = "red"
+        elif cash < DEFAULT_CASH_WARNING:
+            status = "yellow"
+        else:
+            status = "green"
+
+        forecast.append({
+            "date": d.strftime("%Y-%m-%d"),
+            "day_offset": day_offset,
+            "predicted_cash": round(cash, 2),
+            "lower_bound": round(max(lower, 0), 2),
+            "upper_bound": round(upper, 2),
+            "alert_status": status,
+        })
+
+    # 整体预警
+    min_cash = min(f["predicted_cash"] for f in forecast)
+    min_date = next(f["date"] for f in forecast if f["predicted_cash"] == min_cash)
+
+    suggestions = []
+    if min_cash < DEFAULT_CASH_CRITICAL:
+        suggestions.append({
+            "type": "critical",
+            "message": f"预计{min_date}现金余额降至{min_cash:.1f}万，低于警戒线{DEFAULT_CASH_CRITICAL}万",
+            "actions": [
+                "立即催收大额应收账款",
+                "暂停非必要支出",
+                "准备短期融资安排",
+            ]
+        })
+    elif min_cash < DEFAULT_CASH_WARNING:
+        suggestions.append({
+            "type": "warning",
+            "message": f"预计{min_date}现金余额降至{min_cash:.1f}万，低于关注线{DEFAULT_CASH_WARNING}万",
+            "actions": [
+                "加快应收账款回款",
+                "控制采购付款节奏",
+                "评估短期现金流压力",
+            ]
+        })
+
+    return {
+        "entity_id": entity_id,
+        "base_cash": round(base_cash, 2),
+        "days": days,
+        "forecast": forecast,
+        "min_cash": round(min_cash, 2),
+        "min_cash_date": min_date,
+        "trends": {
+            "operating_cash_flow_trend_pct": round(ocf_trend, 2),
+            "receivables_trend_pct": round(ar_trend, 2),
+            "payables_trend_pct": round(ap_trend, 2),
+        },
+        "suggestions": suggestions,
+    }
+
+
+def save_forecast_to_db(entity_id: int, forecast_data: dict, db: Session):
+    """将预测结果保存到数据库"""
+    from app.models import CashForecast
+    for f in forecast_data["forecast"]:
+        forecast_date = datetime.strptime(f["date"], "%Y-%m-%d")
+        cf = CashForecast(
+            entity_id=entity_id,
+            forecast_date=forecast_date,
+            predicted_cash=f["predicted_cash"],
+            lower_bound=f["lower_bound"],
+            upper_bound=f["upper_bound"],
+            alert_status=f["alert_status"],
+        )
+        db.add(cf)
+    db.commit()
+
+
+def calculate_accuracy(entity_id: int, db: Session) -> list:
+    """计算预测准确率 — 对比上期预测 vs 本期实际"""
+    from app.models import CashForecast, KPIDefinition, KPIValue
+
+    # 获取实体最近的预测
+    forecasts = db.query(CashForecast).filter(
+        CashForecast.entity_id == entity_id,
+    ).order_by(CashForecast.forecast_date.desc()).limit(90).all()
+
+    # 获取实际的现金余额KPI值
+    cash_kpi = db.query(KPIDefinition).filter(
+        KPIDefinition.kpi_code == KPI_CODES["cash_balance"],
+        KPIDefinition.entity_id == entity_id,
+    ).first()
+
+    if not cash_kpi or not forecasts:
+        return []
+
+    actuals = db.query(KPIValue).filter(
+        KPIValue.kpi_id == cash_kpi.id,
+        KPIValue.actual_value.isnot(None),
+    ).order_by(KPIValue.period.desc()).limit(12).all()
+
+    actual_map = {}
+    for a in actuals:
+        try:
+            # period like "2026-07" -> month approx
+            actual_map[a.period] = a.actual_value
+        except:
+            pass
+
+    # 按月汇总预测值和实际值，计算准确率
+    from collections import defaultdict
+    monthly_forecast = defaultdict(list)
+    for f in forecasts:
+        month_key = f.forecast_date.strftime("%Y-%m")
+        monthly_forecast[month_key].append(f.predicted_cash)
+
+    results = []
+    for month, f_vals in sorted(monthly_forecast.items()):
+        if month in actual_map:
+            f_avg = sum(f_vals) / len(f_vals)
+            a_val = actual_map[month]
+            mae = abs(f_avg - a_val)
+            mape = abs((f_avg - a_val) / max(abs(a_val), 1)) * 100
+            results.append({
+                "period": month,
+                "forecast_value": round(f_avg, 2),
+                "actual_value": round(a_val, 2),
+                "mae": round(mae, 2),
+                "mape": round(mape, 2),
+            })
+
+    return results
+
+
+def generate_scenario_suggestion(alert_type: str, kpi_name: str, extra: dict = None) -> dict:
+    """根据预警类型生成情景建议"""
+    suggestions = {
+        "cash_low": {
+            "title": "现金流紧张缓解方案",
+            "description": f"现金余额低于阈值，建议加快应收账款催收、控制支出、评估短期融资。",
+            "actions": [
+                f"催收大额应收账款（预计回款{extra.get('expected_receivables', '待定')}万元）",
+                "暂停非紧急采购和资本性支出",
+                "与供应商协商延长账期",
+                "评估银行短期授信额度",
+            ],
+            "priority": "high",
+        },
+        "cash_critical": {
+            "title": "现金流危机应对方案",
+            "description": f"现金余额接近断流，需立即采取紧急措施。",
+            "actions": [
+                "立即催收所有到期应收账款",
+                "暂停所有非必要支出",
+                "紧急联系银行安排短期贷款",
+                "评估资产变现可能性",
+            ],
+            "priority": "high",
+        },
+        "cost_high": {
+            "title": "成本管控优化方案",
+            "description": f"成本率异常偏高，建议进行成本结构分析和优化。",
+            "actions": [
+                "逐项分析成本构成，识别异常项",
+                "与供应商重新谈判采购价格",
+                "评估流程优化降本空间",
+                "建立费用审批红线上限",
+            ],
+            "priority": "medium",
+        },
+        "revenue_drop": {
+            "title": "收入下滑应对方案",
+            "description": f"收入出现下滑趋势，建议分析原因并制定恢复计划。",
+            "actions": [
+                "分析收入下滑原因（客户流失/价格战/需求变化）",
+                "制定客户留存和挽回计划",
+                "评估新产品/新市场机会",
+                "优化销售激励政策",
+            ],
+            "priority": "high",
+        },
+    }
+
+    sug = suggestions.get(alert_type, {
+        "title": "改善建议",
+        "description": "根据预警情况制定改善措施。",
+        "actions": ["分析预警原因", "制定改善计划", "跟踪执行效果"],
+        "priority": "medium",
+    })
+
+    if extra:
+        sug["extra"] = extra
+    return sug