Deep Learning-Based Energy Load Forecasting Using a Hybrid TCN–LSTM Model with Attention Mechanism

Abstract

Accurate energy load forecasting is a fundamental requirement for ensuring the reliable operation, economic efficiency, and sustainability of modern power systems. However, electricity consumption patterns are inherently complex, exhibiting nonlinear behavior and strong seasonal variations, which pose significant challenges for traditional forecasting methods. Conventional statistical approaches often fail to adequately capture these dynamics, resulting in limited prediction accuracy. In this study, a hybrid deep learning framework is proposed that combines Temporal Convolutional Networks (TCN), Long Short-Term Memory (LSTM) networks, and an attention mechanism to effectively model both short-term fluctuations and long-term dependencies in electricity load data. The TCN component is utilized to extract local temporal features through dilated causal convolutions, while the LSTM network captures long-range sequential dependencies. The attention mechanism further enhances the model by selectively focusing on the most relevant time steps, thereby improving both prediction accuracy and interpretability. The proposed model is evaluated using the UCI Electricity Load Diagrams dataset (2011–2014), which contains high-resolution electricity consumption data from 370 customers. Experimental results demonstrate that the hybrid model achieves superior performance, attaining an RMSE of 2.9 and MAE of 2.2, significantly outperforming baseline models including ARIMA, SVR, and standalone deep learning approaches. The model shows an improvement of approximately 44% over ARIMA, 37% over SVR, and 23% over conventional LSTM models, highlighting its effectiveness in capturing complex temporal patterns. The results confirm that the proposed hybrid architecture provides a robust and scalable solution for energy load forecasting. Its ability to accurately model multi-scale temporal dependencies makes it particularly suitable for real-world smart grid applications, supporting improved demand response strategies and efficient energy management.