A Cyclical Forecasting Method for Electricity Load Based on Multiscale Decomposition and Integrated Learning

Résumé fourni par la source

Conventional cyclic load forecasting hierarchies are generally independent, with low forecasting efficiency, resulting in increased forecasting errors. In this regard, we propose to design and analyze a cyclic load forecasting method based on multiscale decomposition and integrated learning. According to the current forecasting demand, data preprocessing and load characteristic decomposition are carried out first, the multiorder form is adopted to strengthen the forecasting efficiency, the integrated multi-order forecasting hierarchy is set up, the multiscale decomposition+integrated learning load cyclic forecasting model is constructed, and the stochastic difference kernel checking is used to realize the cyclic forecasting process. The results show that compared with the load forecasting method of load distribution DC microgrid, the short-term power load forecasting method of two-stage attention mechanism and gated recurrent unit network, the absolute error of the designed multiscale decomposition and integrated learning power load cyclic forecasting method is smaller, which indicates that the designed power load cyclic forecasting method is real, reliable, and covers the loads in a more controlled way with the support of the multiscale decomposition and integrated learning. It shows that with the support of multiscale decomposition and integrated learning, the designed method is real and reliable, with strong controllability and higher prediction accuracy, which is of practical significance.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.