Load Prediction Based on TimesNet and visual Attention Mechanism

Accurate prediction of power load is of great significance to optimize production process, reduce production cost and ensure stable operation of regional power grid.This paper proposes an innovative prediction method that combines the TimesNet architecture with visual attention mechanism for shortterm power load prediction. This method improves prediction accuracy by enhancing modeling capabilities at the periodic level, effectively capturing load fluctuation characteristics and their influencing factors. First, a framework based on TimesNet is constructed to achieve multi-period decomposition and parallel reconstruction of time series, accurately capturing fluctuation characteristics within and between periods. Subsequently, a decoding network module based on visual attention mechanism is proposed, which can accurately mine and decode prediction features of load across multiple periodic levels. Verified through real datasets from multiple cities in China, this method demonstrates higher prediction accuracy and stronger generalization ability compared to traditional methods, providing an efficient and intelligent solution for power load prediction.