A Comprehensive Analysis of a Hybrid Deep Learning Model for Midterm Electric Load Forecasting

In the realm of time series forecasting, particularly for midterm electric load predictions, hybrid deep learning models have shown promising results. This paper introduces a thorough implementation and validation of the established ETS+RD-LSTM model. Leveraging the synergy of Exponential Smoothing (ETS) and Residual Dilated Long Short-Term Memory (RD-LSTM), the model was applied to monthly electricity demand datasets from 10 different countries. The validation process employed was the walk-forward validation method. When juxtaposed with traditional models like ARIMA, MLP, XGBoost, ETS, kNN, and LSTM, the ETS+RD-LSTM model demonstrated superior forecasting accuracy and robustness. An in-depth performance analysis offers insights into the strengths and nuances of the model's prediction capabilities. Through this research, we aim to contribute to the growing body of knowledge on advanced hybrid forecasting models and their practical applications in the energy sector.