MSTM: Multiscale Spatio-Temporal Modeling for Power Grid Line Loss Prediction

The global shift toward renewable energy has accelerated during the past decade, driven by the pursuit of a sustainable and cleaner future. Technological progress has significantly increased the efficiency of renewable energy systems, enhancing their competitive position of these renewable systems relative to traditional fossil fuels. However, the inherent variability and unpredictability of renewable sources, such as solar and wind, present considerable challenges to maintaining the stability and efficiency of power grids. This paper proposes a multiscale spatio-temporal model (MSTM) to improve power line loss predictions, optimize power dispatch, reduce energy waste, and enhance grid stability and efficiency. MSTM leverages multi-level and multiscale modeling of grid topology and designs a deep learning model to extract and fuse spatio-temporal features. Comparative studies demonstrate that MSTM outperforms existing methods regarding inline loss prediction accuracy. Furthermore, through visualization techniques, we investigate and discuss the impact of temporal and topological characteristics on line loss prediction outcomes, providing a deeper insight into the factors influencing line loss in power networks. These findings underscore the potential of advanced spatio-temporal analysis to enhance renewable energy integration.