Long Term Forecasting Method for Microgrid Multi Load Considering Seasonal Influencing Factors

With the increasing role of renewable energy in power grid construction and the diversification of terminal power demand, accurate multivariate load forecasting is becoming more and more important for power system optimal dispatching and power grid stability. Therefore, this paper proposes a forecasting model based on MMoE-RNN-Informer considering seasonal factors. First of all, the multi-task learning model MMoE is used to learn the coupling characteristics of the data characteristics of each load. Secondly, the neural network RNN is used to deal with the sequence data to improve the utilization rate of historical data information. Finally, the weighted summation of the output of the expert network is input into the informer forecasting model of each load to realize the comprehensive seasonal multivariate load long-term and short-term accurate forecasting. The simulation results show that taking the electricity, heating, and cooling load data of the 2019 modeling competition as an example, compared with the other two typical correlation prediction methods, the prediction accuracy of the overall model WMAPE is improved by 2.97% in terms of short-term prediction accuracy, 2.53% in terms of long-term prediction.