Hybrid Attention-Enhanced Deep Learning for Accurate Hourly Energy Consumption Forecasting

Abstract

Accurate forecasting of hourly energy consumption is essential for optimizing energy distribution, ensuring grid stability, and informing policy decisions. In this study, we propose a novel hybrid deep learning model that integrates attention mechanisms with long short-term memory for forecasting hourly electricity consumption. The model is trained and tested using a PJME_MW dataset, spanning from December 31, 2002, to January 2, 2018. The model was evaluated using a set of evaluation metrics R-squared score, Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Deviation (MAD), and Mean Absolute Error (MAE). The models are compared with established deep learning architectures such as ResNet, TCN, LSTM, RNN, and Attention CNN. The Comparative analysis demonstrates superior forecasting performance. The results showed that the proposed model outperformed all other models, it achieved the best accuracy with RMS, MAE, and R2 Score of 0.012, 0.007, and 0.992 respectively, which validates the effectiveness of our approach in enhancing prediction accuracy for energy consumption. The source code is publicly accessible at https://github.com/Hourly-Energy-Consumption.