PConvLSTM: an effective parallel ConvLSTM-based model for short-term electricity load forecasting

Short-term load forecasting (STLF) poses challenges for utility grid systems (UGS) due to unpredictable factors, hindering accurate electricity demand predictions. Despite difficulties, enhancing the STLF technology for accuracy is crucial for maintaining equilibrium, averting outages, and optimizing efficiency amid evolving technological landscapes. The work presented in the manuscript aims to address these challenges by introducing a novel STLF framework that combines a simple feature processing technique and a parallel ConvLSTM network to improve the STLF accuracy. In contrast to most existing models that use 1DCNN, our approach uses ConvLSTM architecture that processes 2D features and effectively captures both temporal and spatial patterns in the time-series load data. The proposed PConvLSTM model was evaluated against various benchmark and state-of-the-art models using publicly available power grid datasets from Malaysia and Tetouan. The results of this evaluative experiment demonstrate that the proposed model consistently outperformed other competing models across various performance metrics such as MAE, MAPE, RMSE, and R². Specifically, in direct comparison with an existing model, our PConvLSTM model demonstrated superior performance, achieving an MAE of 0.021, a MAPE of 1.94, and an impressive R² score of 99.29%. To substantiate these findings, we statistically validate our results using the Friedman and Nemenyi post hoc tests.