Short Term Load Forecasting Of Power Users Based On Deep Learning

Short-term load forecasting refers to the technology of forecasting the load in the future from one day to one week.For the short-term forecasting of power load,it is necessary to take the historical data of the load as the basis,and consider the meteorological conditions,date types,policy guidance and other factors to establish the forecasting model and forecast the future time load value.At present,the research focus is mainly on the use of Deep Learning algorithms to predict short-term load.Relevant scholars explored the influence of various factors on the load value.The current research difficulty is to construct the be st prediction model by selecting the super parameters in the deep learning model.Based on Convolutional Neural Networks(CNN)and Gated Recurrent Unit(GRU),th is paper studied the above-mentioned problems,and the main work and results are as follows:(1)The factors affecting short-term load prediction and the methods of data preprocessing are studied.First,eight months of load data were obtained using smart meters.Secondly,the abnormal data were identified and corrected,the missing load value was filled,and the revised data was normalized.Finally,the influen ce of meteorological factors and date factors on the variation of load value is discussed,and the influence of super parameters in deep learning network on the prediction accuracy is discussed.(2)Differential Evolution-Convolutional Neural Networks-Gated Recurrent Unit optimized based on Differential Evolution algorithm was studied for short-term load forecasting.Firstly,the network characteristics of CNN network and GRU network are discussed.Secondly,this paper combines the advantages of CNN n etwork and GRU network,proposes CNN-GRU network,and introduces the network structure of CNN-GRU network.Finally,the algorithm principle of differential optimization algorithm and the algorithm flow of DE-CNN-GRU network are introduced.(3)Analysis of calculation examples.The load data are used to evaluate the network models of CNN network,GRU network and DE-CNN-GRU network respectively.The results show that DE-CNN-GRU network has higher accuracy and better generalization ability than CNN network and GRU network.