An ensemble deep learning framework for energy demand forecasting using genetic algorithm-based feature selection.

Area of Science:

• Energy Systems
• Artificial Intelligence
• Data Science

Background:

• Accurate energy demand forecasting is essential for efficient energy management and planning.
• Machine learning models have advanced significantly but feature selection remains a challenge.

Purpose of the Study:

• To propose an ensemble approach for optimizing feature selection in energy demand forecasting.
• To enhance prediction accuracy and robustness using a genetic algorithm and multiple forecasting models.

Main Methods:

• An ensemble approach integrating a genetic algorithm for feature selection with LSTM, BiLSTM, and GRU models.
• Stacking ensemble technique to combine base learner predictions.
• Dataset divided into weekday and weekend subsets for detailed analysis.
• Ten simulations and Wilcoxon Signed Rank Test for reliability.

Main Results:

• Achieved high accuracy with RMSE of 130.6, MAPE of 0.38%, and MAE of 99.41 for weekday energy demand forecasting.
• Maintained strong performance for weekend predictions with RMSE of 137.41, MAPE of 0.42%, and MAE of 105.67.
• Demonstrated exceptional precision and robustness in energy demand predictions.

Conclusions:

• The proposed ensemble model effectively optimizes feature selection for improved energy demand forecasting.
• This research offers valuable insights for energy analysts and contributes to advanced forecasting methods.