ML Forecasting

Kaggle ingestion → augmented dataset → scikit-learn training → formula/ml/hybrid forecasting.

Kaggle· solar-power-generation-plantscikit-learn· HistGradientBoostingRegressorpvlib· GHI→MW physics conversion

Formulas used

ML target: irradiance predictionMODEL_LEARNED

model.predict(features) → GHI_pred
# HistGradientBoostingRegressor
# Features: temp, module_temp, irradiance,
#            wind_speed, humidity, hour, month

Irradiance → generation (pvlib)OFFICIAL_SOURCE · pvlib

Erbs(ghi, zenith) → DNI, DHI
POA = get_total_irradiance(...)
t_cell = faiman(poa, temp, wind)
DC = pvwatts_dc(poa, t_cell, pdc0, γ)
AC = pvwatts_inverter(DC, η=0.96)
MW = min(AC / 1e6, capacity)

MAE (Mean Absolute Error)OFFICIAL_SOURCE · sklearn

MAE = (1/n) Σ |y_true − y_pred|

R² (Coefficient of determination)OFFICIAL_SOURCE · sklearn

R² = 1 − SS_res / SS_tot
SS_res = Σ (y_true − y_pred)²
SS_tot = Σ (y_true − mean)²

RMSE (Root Mean Square Error)OFFICIAL_SOURCE · sklearn

RMSE = √( (1/n) Σ (y_true − y_pred)² )

MAPE (Mean Abs % Error)OFFICIAL_SOURCE · sklearn

MAPE = (100/n) Σ |y_true − y_pred| / |y_true|

Provenance: MODEL_LEARNED + OFFICIAL_SOURCE — The ML model is trained on real Kaggle solar plant data; generation is derived via pvlib physics, not learned directly. Model card stored in models/metadata/.

Checking backend…