An Extended Frequency-Improved Legendre Memory Model for Enhanced Long-Term Electricity Load Forecasting

Long-term electricity load forecasting is crucial for energy conservation, grid planning, and reducing carbon emissions by enabling optimal resource allocation and efficient energy utilization. However, forecasting the highly fluctuating loads in a large electrical power grid presents significant challenges due to the variability and complexity of individual load patterns across buses. Traditional models primarily focus on establishing temporal dependencies, often neglecting critical relationships between feature variables. This study introduces a novel approach that integrates de Bruijn Graphs (dBGs) with state-of-the-art time-series models to enhance predictive capabilities. By leveraging the unique structural properties of dBGs, the proposed framework improves the representation of sequential dependencies in power grid data. Advanced graph encoding techniques are utilized to extract meaningful features from dBGs that are often overlooked by traditional methods. Four enhanced architectures—FiLMdBG, iTransformerdBG, TimesNetdBG, and DLineardBG—are developed and evaluated on the Texas 2,000-bus test system across multiple forecasting horizons. The results demonstrate that dBG-integrated models significantly outperform their conventional counterparts, delivering superior accuracy in both short and long-term electricity load forecasting. These findings underscore the potential of dBGs as a transformative tool for advancing power grid management and enabling more sustainable and efficient energy systems.