BiLSTM-CNN with Bayesian Optimization for Accurate Long-Term Load Forecasting: Cross-Regional Insights from Morocco and Spain

Accurate long-term electricity demand forecasting (LTLF) is critical for strategic planning, particularly in the context of escalating climate issues and the intricacies of energy systems. This research proffers a pioneering hybrid deep learning model that integrates Bidirectional Long Short-Term Memory (BiLSTM) networks and Convolutional Neural Networks (CNN), which is optimised using Bayesian Optimisation (BO), with a view to enhancing forecast reliability across diverse energy landscapes. Utilising hourly data from Morocco and Spain in 2017, the model captures significant seasonal, meteorological, and socioeconomic factors that influence power usage. The integration of advanced feature engineering techniques, including lag features and rolling statistical windows, enhances temporal representation, while CNN layers facilitate the extraction of spatial relationships. The model provides reliable 30-day forecasts, validated with MAE, RMSE, and R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> metrics. The model demonstrates higher accuracy in Spain (RMSE: 798.03 kW, R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>: 0.9693) and performs well in Morocco (RMSE: 1836.91 kW, R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>: 0.9324), thus demonstrating its versatility. This methodology provides a scalable solution for utilities and regulators looking to address long-term demand uncertainties and promote renewable integration.