A Novel Framework to Enhancing Long-Term Time Series Forecasting in the Electrical Domain

Accurate long-term forecasting of electrical time series is crucial for the reliable operation and maintenance of power systems. Traditional statistical models often fall short in capturing the complex temporal dependencies and nonlinear patterns inherent in electrical data. This paper proposes an enhanced PatchTST-based framework for long-term electrical time series forecasting, integrating systematic feature extraction, statistical evaluation, and Bayesian-optimised patch selection. The proposed approach improves upon existing transformer models by more effectively capturing complex temporal patterns while maintaining computational efficiency. Evaluated on the Electricity and ETT datasets, the framework outperforms state-of-the-art methods across multiple benchmarks, achieving up to 22.9% lower MSE. The results indicate that integrating domain knowledge with advanced transformer architectures can significantly enhance long-term prediction performance in electrical time series. This study contributes to the advancement of forecasting techniques, offering practical implications for smart grid management and condition monitoring.