<![CDATA[Long-term electricity load forecasting plays an important role in ensuring system reliability, optimizing energy management, and making operational plans in face of continuously rising electricity demands. This study suggests a complete deep learning method for univariate forecasting of future electricity loads based on climatology and electricity consumption data for the period between 2019 and 2023. The initial dataset was cleaned, normalized, and partitioned chronologically into train/test datasets. Four train/test split cases (20/80, 40/60, 60/40, 80/20) were considered to explore the impact of different levels of historical data availability on the performance of the suggested framework from data-poor to data-rich situations. Then five deep learning structures (CNN, RNN, GRU, LSTM, and BiLSTM) were trained and tuned by using three different hyperparameter optimization methods. Grid Search provided an extensive exploration of parameter space to obtain solid baseline configurations for the considered neural networks, Random Search allowed efficient sampling of the search space to find high-quality deep learning models with lower computational expenses, and Particle Swarm Optimization (PSO) enabled adaptive optimization of near-optimal solutions via population-based optimization technique. Forecasting models were assessed by means of Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE) metrics. Technical novelty of the study is related to the consideration of the impact of various hyperparameter optimization approaches on the quality and robustness of predictions provided by several deep learning architectures depending on the degree of historical data availability. Experiment results showed that the CNN tuned with the help of Random Search gave the best forecasting results in case of abundant training data (80/20 split) with RMSE=22.13. In case of scarcity of training samples (20/80 split) CNN tuned by Grid Search and PSO provided stable forecasts with MAPE≈0.077 and RMSE=37.56 with efficient reduction of prediction deviation.]]>
