De Bruijn Graph-Enhanced Time Series Models for Electricity Load Forecasting

Effective electricity load prediction enables grid operators to design optimal generation schedules and energy dispatch strategies that minimize risks from demand fluctuations. However, predicting highly volatile loads in large-scale power grids remains a complex challenge due to the dynamic nature of individual bus-level consumption patterns. Traditional forecasting methods primarily focus on temporal trends and often overlook interdependencies between influencing factors, which limits their ability to capture load variations. To address these challenges, this study introduces a methodology that integrates de Bruijn Graphs (dBGs) with advanced time-series forecasting models. By leveraging structural properties, the framework enhances the modeling of sequential dependencies within power grid data. Advanced graph encoding techniques extract salient features from dBGs and help identify overlooked patterns. This study develops four enhanced architectures—FiLMdBG, iTransformerdBG, TimesNetdBG, and DLineardBG—evaluated on the Texas 2,000-bus test system across various forecasting horizons. Empirical results demonstrate that dBG-enhanced models outperform traditional approaches and achieve superior accuracy in both short-term and long-term electricity load forecasting. These findings highlight the potential of dBGs as a key tool for improving power grid management and advancing sustainable energy systems.