File size: 16,079 Bytes
7acfdcf
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
"""
GPU-accelerated load forecasting using STL decomposition + XGBoost.
Implements the forecasting methodology from the research paper.
"""

import pandas as pd
import numpy as np
from typing import Tuple, Optional, Dict
import logging
from pathlib import Path
import pickle

# Time series decomposition
from statsmodels.tsa.seasonal import STL

# Linear regression for trend extrapolation
from sklearn.linear_model import LinearRegression

# XGBoost with GPU support
import xgboost as xgb

# Visualization
import matplotlib.pyplot as plt
import seaborn as sns

from .config import (
    model_config,
    gpu_config,
    MODELS_DIR,
    RESULTS_DIR
)

logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

sns.set_style("whitegrid")


class STLXGBoostForecaster:
    """
    Two-stage forecasting model:
    1. STL decomposition to separate trend, seasonal, and residual components
    2. XGBoost regression on residuals with GPU acceleration
    """

    def __init__(
        self,
        seasonal_period: int = None,
        trend_window: int = None,
        use_gpu: bool = True
    ):
        """
        Initialize forecaster

        Args:
            seasonal_period: Period for seasonal decomposition (default: 24 hours)
            trend_window: Window for trend extraction (default: 168 hours = 1 week)
            use_gpu: Whether to use GPU acceleration for XGBoost
        """
        self.seasonal_period = seasonal_period or model_config.stl_seasonal_period
        self.trend_window = trend_window or model_config.stl_trend_window
        self.use_gpu = use_gpu and gpu_config.use_gpu

        # Model components
        self.stl_model = None
        self.xgb_model = None
        self.trend_component = None
        self.seasonal_component = None
        self.trend_regression_model = None  # Linear regression for trend extrapolation
        self.n_trend_samples = None  # Number of trend samples for time index

        # Metadata
        self.feature_names = None
        self.train_stats = {}

    def decompose_time_series(
        self,
        y_train: pd.Series,
        timestamps: pd.Series = None
    ) -> Dict[str, np.ndarray]:
        """
        Perform STL decomposition

        Args:
            y_train: Training time series
            timestamps: Optional timestamps for plotting

        Returns:
            Dictionary with trend, seasonal, and residual components
        """
        logger.info("Performing STL decomposition...")

        # Ensure we have enough data
        min_required = 2 * self.seasonal_period
        if len(y_train) < min_required:
            raise ValueError(f"Need at least {min_required} samples for STL decomposition")

        # Validate STL parameters: trend must be odd and > period
        trend_window = self.trend_window
        if trend_window <= self.seasonal_period:
            trend_window = self.seasonal_period + 2
            logger.warning(f"trend_window ({self.trend_window}) must be > period ({self.seasonal_period}). Using {trend_window}")

        if trend_window % 2 == 0:
            trend_window += 1
            logger.warning(f"trend_window must be odd. Using {trend_window}")

        # Reset index to ensure proper time series (STL needs integer index)
        y_train_reset = y_train.reset_index(drop=True)

        # Fit STL - use 'period' parameter explicitly
        self.stl_model = STL(
            y_train_reset,
            period=self.seasonal_period,  # Period for seasonal decomposition
            trend=trend_window,  # Must be odd and > period
            robust=True  # Robust to outliers
        )

        result = self.stl_model.fit()

        # Extract components
        components = {
            'trend': result.trend.values,
            'seasonal': result.seasonal.values,
            'residual': result.resid.values,
            'observed': y_train.values
        }

        # Store for later use
        self.trend_component = components['trend']
        self.seasonal_component = components['seasonal']

        # Log statistics
        logger.info(f"STL Decomposition Statistics:")
        logger.info(f"  Trend range: [{components['trend'].min():.2f}, {components['trend'].max():.2f}]")
        logger.info(f"  Seasonal range: [{components['seasonal'].min():.2f}, {components['seasonal'].max():.2f}]")
        logger.info(f"  Residual std: {components['residual'].std():.2f}")

        return components

    def fit(
        self,
        X_train: pd.DataFrame,
        y_train: pd.Series,
        X_val: Optional[pd.DataFrame] = None,
        y_val: Optional[pd.Series] = None,
        plot_decomposition: bool = True
    ):
        """
        Fit the complete forecasting model

        Args:
            X_train: Training features
            y_train: Training target
            X_val: Validation features (optional)
            y_val: Validation target (optional)
            plot_decomposition: Whether to plot STL decomposition
        """
        logger.info("Fitting STL-XGBoost forecaster...")

        # Store feature names
        self.feature_names = X_train.columns.tolist()

        # Step 1: STL Decomposition
        components = self.decompose_time_series(y_train)

        # Step 1.5: Fit trend regression model for proper extrapolation
        logger.info("Fitting linear regression model for trend extrapolation...")
        self.n_trend_samples = len(self.trend_component)
        time_index = np.arange(self.n_trend_samples).reshape(-1, 1)
        self.trend_regression_model = LinearRegression()
        self.trend_regression_model.fit(time_index, self.trend_component)
        trend_slope = float(self.trend_regression_model.coef_[0])
        trend_intercept = float(self.trend_regression_model.intercept_)
        logger.info(f"  Trend equation: y = {trend_intercept:.2f} + {trend_slope:.6f} * t")

        # Plot decomposition if requested
        if plot_decomposition:
            self._plot_decomposition(components)

        # Step 2: Train XGBoost on residuals
        logger.info("Training XGBoost model on residuals...")

        # Prepare XGBoost parameters (XGBoost 3.1+ compatible with regularization)
        params = {
            'objective': 'reg:squarederror',
            'max_depth': model_config.xgb_max_depth,
            'learning_rate': model_config.xgb_learning_rate,
            'subsample': model_config.xgb_subsample,  # Subsample 80% to prevent overfitting
            'colsample_bytree': model_config.xgb_colsample_bytree,  # Use 80% of features
            'reg_alpha': model_config.xgb_reg_alpha,  # L1 regularization
            'reg_lambda': model_config.xgb_reg_lambda,  # L2 regularization
            'tree_method': 'hist',  # Use 'hist' for both CPU and GPU (device parameter controls GPU usage)
            'device': 'cuda' if self.use_gpu else 'cpu',  # XGBoost 3.1+ uses 'device' instead of 'gpu_id'
            'random_state': model_config.random_seed,
            'verbosity': 1
        }

        # Create DMatrix for efficient training
        dtrain = xgb.DMatrix(X_train, label=components['residual'])

        # Prepare validation set if provided
        eval_set = []
        if X_val is not None and y_val is not None:
            # Decompose validation target using trend regression
            # Validation trend continues from where training ended
            val_time_index = np.arange(self.n_trend_samples, self.n_trend_samples + len(y_val)).reshape(-1, 1)
            val_trend = self.trend_regression_model.predict(val_time_index)

            val_seasonal = np.tile(
                self.seasonal_component[:self.seasonal_period],
                len(y_val) // self.seasonal_period + 1
            )[:len(y_val)]
            val_residual = y_val.values - val_trend - val_seasonal

            dval = xgb.DMatrix(X_val, label=val_residual)
            eval_set = [(dtrain, 'train'), (dval, 'val')]

        # Train model
        evals_result = {}
        self.xgb_model = xgb.train(
            params,
            dtrain,
            num_boost_round=model_config.xgb_n_estimators,
            evals=eval_set if eval_set else [(dtrain, 'train')],
            early_stopping_rounds=model_config.xgb_early_stopping_rounds,
            evals_result=evals_result,
            verbose_eval=50
        )

        # Store training statistics
        self.train_stats = {
            'train_rmse': evals_result['train']['rmse'][-1] if 'train' in evals_result else None,
            'val_rmse': evals_result['val']['rmse'][-1] if 'val' in evals_result else None,
            'best_iteration': self.xgb_model.best_iteration,
            'n_features': len(self.feature_names)
        }

        logger.info(f"Training complete. Best iteration: {self.xgb_model.best_iteration}")
        if self.train_stats['val_rmse']:
            logger.info(f"Validation RMSE: {self.train_stats['val_rmse']:.2f}")

    def predict(
        self,
        X: pd.DataFrame,
        forecast_steps: int = None
    ) -> np.ndarray:
        """
        Make predictions

        Args:
            X: Feature matrix
            forecast_steps: Number of steps to forecast (for extrapolating trend/seasonal)

        Returns:
            Predictions array
        """
        if self.xgb_model is None:
            raise ValueError("Model must be fitted before prediction")

        # Predict residuals
        dtest = xgb.DMatrix(X)
        residual_pred = self.xgb_model.predict(dtest)

        # Reconstruct trend and seasonal components
        n_pred = len(X)

        # Extrapolate trend using LINEAR REGRESSION (proper extrapolation)
        # Time indices continue from where training ended
        time_index_pred = np.arange(self.n_trend_samples, self.n_trend_samples + n_pred).reshape(-1, 1)
        trend_pred = self.trend_regression_model.predict(time_index_pred)

        # Repeat seasonal pattern
        seasonal_pred = np.tile(
            self.seasonal_component[:self.seasonal_period],
            n_pred // self.seasonal_period + 1
        )[:n_pred]

        # Combine components
        y_pred = trend_pred + seasonal_pred + residual_pred

        return y_pred

    def evaluate(
        self,
        X_test: pd.DataFrame,
        y_test: pd.Series,
        plot: bool = True
    ) -> Dict[str, float]:
        """
        Evaluate model performance

        Args:
            X_test: Test features
            y_test: Test target
            plot: Whether to plot predictions

        Returns:
            Dictionary of evaluation metrics
        """
        logger.info("Evaluating model on test set...")

        # Make predictions
        y_pred = self.predict(X_test)

        # Calculate metrics
        from sklearn.metrics import mean_absolute_error, mean_squared_error, r2_score

        mae = mean_absolute_error(y_test, y_pred)
        rmse = np.sqrt(mean_squared_error(y_test, y_pred))
        mape = np.mean(np.abs((y_test.values - y_pred) / y_test.values)) * 100
        r2 = r2_score(y_test, y_pred)

        metrics = {
            'MAE': mae,
            'RMSE': rmse,
            'MAPE': mape,
            'R2': r2
        }

        logger.info("Test Set Metrics:")
        for metric, value in metrics.items():
            logger.info(f"  {metric}: {value:.4f}")

        # Plot if requested
        if plot:
            self._plot_predictions(y_test.values, y_pred)

        return metrics

    def _plot_decomposition(self, components: Dict[str, np.ndarray]):
        """Plot STL decomposition components"""
        fig, axes = plt.subplots(4, 1, figsize=(15, 10))

        components_to_plot = ['observed', 'trend', 'seasonal', 'residual']
        titles = ['Observed', 'Trend', 'Seasonal', 'Residual']

        for ax, comp_name, title in zip(axes, components_to_plot, titles):
            ax.plot(components[comp_name], linewidth=1)
            ax.set_title(title, fontsize=12, fontweight='bold')
            ax.set_ylabel('MW')
            ax.grid(True, alpha=0.3)

        axes[-1].set_xlabel('Time Index')
        plt.tight_layout()

        # Save plot
        plot_path = RESULTS_DIR / 'stl_decomposition.png'
        plt.savefig(plot_path, dpi=300, bbox_inches='tight')
        logger.info(f"Saved decomposition plot to {plot_path}")
        plt.close()

    def _plot_predictions(self, y_true: np.ndarray, y_pred: np.ndarray):
        """Plot actual vs predicted values"""
        fig, axes = plt.subplots(2, 1, figsize=(15, 10))

        # Time series plot (first 7 days)
        n_plot = min(24 * 7, len(y_true))  # 1 week
        axes[0].plot(y_true[:n_plot], label='Actual', linewidth=2, alpha=0.7)
        axes[0].plot(y_pred[:n_plot], label='Predicted', linewidth=2, alpha=0.7)
        axes[0].set_title('Load Forecast vs Actual (First Week)', fontsize=12, fontweight='bold')
        axes[0].set_xlabel('Hour')
        axes[0].set_ylabel('Load (MW)')
        axes[0].legend()
        axes[0].grid(True, alpha=0.3)

        # Scatter plot
        axes[1].scatter(y_true, y_pred, alpha=0.3, s=10)
        axes[1].plot([y_true.min(), y_true.max()],
                    [y_true.min(), y_true.max()],
                    'r--', linewidth=2, label='Perfect Prediction')
        axes[1].set_title('Predicted vs Actual Load', fontsize=12, fontweight='bold')
        axes[1].set_xlabel('Actual Load (MW)')
        axes[1].set_ylabel('Predicted Load (MW)')
        axes[1].legend()
        axes[1].grid(True, alpha=0.3)

        plt.tight_layout()

        # Save plot
        plot_path = RESULTS_DIR / 'forecast_predictions.png'
        plt.savefig(plot_path, dpi=300, bbox_inches='tight')
        logger.info(f"Saved prediction plot to {plot_path}")
        plt.close()

    def get_feature_importance(self, top_n: int = 20) -> pd.DataFrame:
        """
        Get feature importance from XGBoost model

        Args:
            top_n: Number of top features to return

        Returns:
            DataFrame with feature importance scores
        """
        if self.xgb_model is None:
            raise ValueError("Model must be fitted first")

        # Get importance scores
        importance = self.xgb_model.get_score(importance_type='gain')

        # Convert to dataframe
        importance_df = pd.DataFrame([
            {'feature': k, 'importance': v}
            for k, v in importance.items()
        ]).sort_values('importance', ascending=False)

        return importance_df.head(top_n)

    def save_model(self, filename: str = "stl_xgboost_model.pkl"):
        """Save model to disk"""
        filepath = MODELS_DIR / filename

        model_data = {
            'xgb_model': self.xgb_model,
            'trend_component': self.trend_component,
            'seasonal_component': self.seasonal_component,
            'seasonal_period': self.seasonal_period,
            'trend_window': self.trend_window,
            'feature_names': self.feature_names,
            'train_stats': self.train_stats
        }

        with open(filepath, 'wb') as f:
            pickle.dump(model_data, f)

        logger.info(f"Saved model to {filepath}")

    def load_model(self, filename: str = "stl_xgboost_model.pkl"):
        """Load model from disk"""
        filepath = MODELS_DIR / filename

        with open(filepath, 'rb') as f:
            model_data = pickle.load(f)

        self.xgb_model = model_data['xgb_model']
        self.trend_component = model_data['trend_component']
        self.seasonal_component = model_data['seasonal_component']
        self.seasonal_period = model_data['seasonal_period']
        self.trend_window = model_data['trend_window']
        self.feature_names = model_data['feature_names']
        self.train_stats = model_data['train_stats']

        logger.info(f"Loaded model from {filepath}")


if __name__ == "__main__":
    print("="*60)
    print("STL-XGBoost Forecasting Demo")
    print("="*60)

    # This will be run with actual data
    logger.info("This module provides GPU-accelerated load forecasting")
    logger.info("Import and use with preprocessed ERCOT data")