Trustworthy Load Forecasting With Generative AI: A Dual-Attention ConvLSTM and VAE-Based Approach

Increasing urbanization and the global transition toward sustainable, eco-friendly energy systems require efficient and robust energy predictions for smart grids. The inherently unpredictable, volatile, and intermittent nature of energy demand necessitates an accurate short-term load forecasting model to ensure reliable consumer applications. However, conventional deep learning models often struggle to address complex and dynamic load patterns. To address these challenges, this research presents a novel trustworthy GAI-assisted model comprising i) a variational autoencoder that maps raw energy consumption data to extract meaningful and compact features and ii) a deep learning model utilizing a dual attention mechanism with convolutional long short-term memory (DAConvLSTM), that effectively captures the temporal dependencies of the complex load pattern and optimizes forecasting accuracy. The effectiveness and robustness of the proposed model are extensively evaluated using publicly available comprehensive datasets. The results demonstrate the performance of the proposed model, with an overall improvement of 1.45%81.54% in the mean absolute error, 1.92%78.61% in the root mean square error, and 1.55%81.85% in the mean absolute percentage error compared with other baseline methods. The results validate the effectiveness of the proposed model in predicting peak load demand and have practical implications, thereby enhancing the existing knowledge for creating robust energy management in smart grid applications.