Forecast Engine (TypeScript)

TÓM TẮT

forecastEngine.ts (~800+ dòng, 50KB) là frontend forecast orchestrator. Kết hợp BAU baseline (Holt-Winters) + Event Impact (calendar weights) + ML Model Results (from Supabase) để tạo daily forecast.

File: src/services/modules/forecast/forecastEngine.ts

Sub-modules (15 files)

Module	File	Mô tả
Core Engine	forecastEngine.ts	Main orchestrator (~50KB)
BAU Baseline	baselineCalculator.ts	Holt-Winters exponential smoothing
Exponential Smoothing	exponentialSmoothing.ts	Core HW implementation
Ensemble	ensembleForecaster.ts	Multi-model ensemble weights
Decision Tree	decisionTree.ts	Staffing decision rules
Anomaly Detection	anomalyDetection.ts	Statistical outlier detection
Auto Tuning	autoTuning.ts	Auto-tune forecast params
Forecast Monitor	forecastMonitor.ts	Accuracy monitoring
Forecast Tuning	forecastTuning.ts	Manual parameter tuning UI
Hierarchical	hierarchicalForecast.ts	Multi-level reconciliation
ML Cache	mlForecastCache.ts	Cache ML results from Supabase
ML Client	mlForecastClient.ts	Fetch ML predictions
Prophet	prophetIntegration.ts	Prophet model integration
Quantile	quantileRegression.ts	Confidence band calculation
Forecast Service	forecast.ts	High-level API

Flow chính: runCombinedForecast()

Output Format

interface ForecastOutput {
  totalVolume: number;
  totalWlu: number;           // WLU-weighted volume
  hourlyDistribution: number[];
  bauVolume: number;          // BAU baseline
  eventVolume: number;        // Event-driven additional volume
  holidayMultiplier: number;  // Holiday adjustment factor
  confidence_p10: number;     // Lower bound
  confidence_p90: number;     // Upper bound
}

THIẾT KẾ CẦN XEM XÉT

Frontend có forecast engine riêng (Holt-Winters) trùng lặp với backend ML. Khi ML accuracy cải thiện, nên chuyển sang ML-primary. Xem ADR-003.

Tài liệu liên quan

• Workforce Engine — Sử dụng output từ forecast engine
• ML Engine (Python) — Backend ML counterpart
• Config Registry — Configs cho forecast params