Hybrid TCN-Based Bi-GRU-LSTM for Enhanced Long-Term Electric Load Forecasting

Abstract

Electric load forecasting plays a vital role in the effective management and planning of energy within power systems.       his research provides an in-depth assessment of several deep learning frameworks for electric load forecasting, concentrating on four specific models: LSTM, GRU-LSTM, Bi-GRU-LSTM, and an innovative TCN-Bi-GRU-LSTM architecture. The models underwent training and testing using a real-world dataset, with input features derived from timestamps to capture temporal trends. The forecasting effectiveness was evaluated over daily, weekly, and monthly timeframes utilizing metrics such as Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE). Although the conventional LSTM and GRU-LSTM models showed constraints in their accuracy for long-term forecasts, the TCN-Bi-GRU-LSTM model successfully leveraged convolutional layers to understand intricate temporal relationships, leading to enhanced performance, especially in monthly predictions. The results suggest that the combination of convolutional layers with recurrent architectures significantly boosts the model's capability to identify long-range patterns, thus enhancing load forecasting precision. This study advances the field of energy forecasting by highlighting the advantages of hybrid architectures in tackling the difficulties associated with long-term load prediction