Table B1 Hyperparameters used in each ML algorithm

M1

PV Count per HH

XGBoost

‘gamma’: 0, ‘alpha’: 12, ‘learning_rate’: 0.027, ‘seed’: 712

‘colsample_bytree’: 0.3, ‘reg_lambda’: 1,’random_state’: 700,

‘n_estimators’: 299, ‘base_score’: 0.29, ‘max_depth’: 7

M2

CATBoost

‘l2_leaf_reg’: 2, ‘learning_rate’: 0.1, ‘depth’: 9, ‘iterations’: 150

M3

LightGBM

‘objective’: ‘regression’, ‘metric’: ‘rmse’,’is_unbalance’: ‘true’,

‘is_training_metric’: ‘true’, ‘boosting’: ‘gbdt’, ‘num_leaves’: 36,

‘feature_fraction’: 0.99, ‘bagging_fraction’: 0.69, ‘bagging_freq’: 4,

‘learning_rate’: 0.01, ‘max_depth’: 15, ‘max_bin’: 23

M4

RandomForest

‘n_estimators’: 19, ‘max_depth’: 150, ‘min_samples_split’: 2,

‘max_features’: “sqrt”,’min_samples_leaf’: 2, ‘random_state’: 531

M5

PV Count per HH + Energy Policy

XGBoost

‘gamma’: 0, ‘alpha’: 5, ‘learning_rate’: 0.05, ‘random_state’: 185,

‘colsample_bytree’: 0.5, ‘reg_lambda’: 0,

‘n_estimators’: 311, ‘base_score’: 0.5, ‘max_depth’: 7, ‘seed’: 855

M6

CATBoost

‘l2_leaf_reg’: 2, ‘learning_rate’: 0.1, ‘depth’: 6, ‘iterations’: 200

M7

LightGBM

‘objective’: ‘regression’, ‘metric’: ‘rmse’,’is_unbalance’: ‘true’,

‘is_training_metric’: ‘true’, ‘boosting’: ‘gbdt’, ‘num_leaves’: 36,

‘feature_fraction’: 0.81, ‘bagging_fraction’: 0.91, ‘bagging_freq’: 20,

‘learning_rate’: 0.021, ‘max_depth’: 14, ‘max_bin’: 23

M8

RandomForest

‘n_estimators’: 700, ‘max_depth’: 150, ‘min_samples_split’: 2,

‘max_features’: “sqrt”,’min_samples_leaf’: 2, ‘random_state’: 372

M9

PV-to-Roof Ratio

XGBoost

‘gamma’: 0, ‘alpha’: 12, ‘learning_rate’: 0.025, ‘seed’:712

‘colsample_bytree’: 0.35, ‘reg_lambda’: 1, ‘random_state’: 789,

‘n_estimators’:300, ‘base_score’: 0.5, ‘max_depth’: 8

M10

CATBoost

‘l2_leaf_reg’: 1, ‘learning_rate’: 0.09, ‘depth’: 10, ‘iterations’: 200

M11

LightGBM

‘objective’: ‘regression’, ‘metric’: ‘rmse’,’is_unbalance’: ‘true’,

‘is_training_metric’: ‘true’, ‘boosting’: ‘gbdt’, ‘num_leaves’: 45,

‘feature_fraction’: 0.25, ‘bagging_fraction’: 0.75, ‘bagging_freq’: 4,

‘learning_rate’: 0.01, ‘max_depth’: 15, ‘max_bin’: 52

M12

RandomForest

‘n_estimators’: 300, ‘max_depth’: 64, ‘min_samples_split’: 3,

‘max_features’: sqrt, ‘min_samples_leaf’: 2, ‘random_state’: 435

M13

PV-to-Roof Ratio + Energy Policy

XGBoost

‘gamma’: 0, ‘alpha’: 5, ‘learning_rate’: 0.05, ‘seed’: 1164

‘colsample_bytree’: 0.5, ‘reg_lambda’: 0,’random_state’: 185,

‘n_estimators’: 500, ‘base_score’: 0.52, ‘max_depth’: 9

M14

CATBoost

‘l2_leaf_reg’: 1, ‘learning_rate’: 0.09, ‘depth’: 6, ‘iterations’: 150

M15

LightGBM

‘objective’: ‘regression’, ‘metric’: ‘rmse’,’is_unbalance’: ‘true’,

‘is_training_metric’: ‘true’, ‘boosting’: ‘gbdt’, ‘num_leaves’: 36,

‘feature_fraction’: 0.34, ‘bagging_fraction’: 0.75, ‘bagging_freq’: 4,

‘learning_rate’: 0.01, ‘max_depth’: 15, ‘max_bin’: 23

M16

RandomForest

‘n_estimators’: 300, ‘max_depth’: 280, ‘min_samples_split’: 2,

‘max_features’: sqrt, ‘min_samples_leaf’: 2, ‘random_state’: 42