Power Forecasting

A comprehensive forecasting that generates 12-hour power forecasts for various assets including PV, BESS, loads, EV chargers, and wind generators.

Features

• Automated Forecast Generation: Generate 12-hour power forecasts every 1-2 hours
• Hourly Forecasts: Generates forecasts at rounded hours (10:00, 11:00, 12:00, etc.)
• Data Validation: Automatically check which devices have sufficient historical data
• Multiple Model Types: Support for Prophet, Moving Average, and Persistence models
• Intelligent Retraining: Periodic model retraining as more data becomes available
• Database-Centric: Store latest forecasts in PostgreSQL
• Asset Type Aware: Tailored forecasting approaches for different asset types

Asset Types and Requirements

Minimum Data Requirements

Different asset types have different minimum data requirements before forecast can be generated:

Asset Type	Minimum Samples (Hourly)	Days	Notes
PV	168	7 days	Strong daily seasonality
WIND	168	7 days	Weather-dependent patterns
BESS	168	7 days	Charge/discharge cycles
AFE	168	7 days	Grid interaction patterns
LOAD	336	14 days	Variable usage patterns
CRITICAL_LOAD	336	14 days	More stable than regular load
UNI_EV	336	14 days	Charging patterns vary by day
BI_EV (V2G)	336	14 days	Complex V2G patterns

Model Selection

The system automatically selects the best model type for each asset:

• PV, WIND: Prophet (strong seasonal patterns) or Moving Average fallback
• BESS, LOAD, EV: Moving Average (pattern-based)
• Others: Moving Average

Model Retraining Strategy

Automatic Retraining Triggers

Models are retrained when:

1. High Priority (immediate):

   • More than 7 days since last training
   • More than 1,000 new data points

2. Medium Priority (soon):

   • More than 3 days since last training
   • More than 500 new data points

3. Low Priority (when convenient):

   • More than 1 day since last training
   • More than 100 new data points

Minimum Retraining Interval

• Models are not retrained more frequently than once per 24 hours
• This prevents excessive computation and allows data to accumulate