Research On Short-term Power Load Forecasting Method Based On GM-LSSVM-MKPCA And RBF Neural Network

Power load is the lifeblood of a country's economic development.Accurate power load forecasting is the basis for power grid departments to carry out power dispatching and formulate power development plans.For example,China's novel coronavirus in 2020,all parts of the country,especially Hubei Province,need to ensure the orderly power supply,also,ensure the normal operation of all hospital equipment and the normal supply of electricity for residents' daily life,so as to ensure that there is no risk.In this case,it can help the power sector to formulate emergency plan and safeguard measures scientifically and effectively by power load forecasting,and power load forecasting is also the experts and scholars,In recent years,power load forecasting is also the focus of experts and scholars.This paper mainly improves the accuracy of power load forecasting by combining improved learning algorithm.The Grey Model(GM)can modify the data by accumulation calculation,it is suitable to deal with random systems with few sample data and significant characteristics.However,the model lacks certain learning ability for power load data and weak information processing ability,which needs further improvement;The Least Squares Support Vector Machine model(LSSVM)can effectively deal with the problems of small samples,local minimum points,nonlinearity and so on.This paper presents a combined forecasting method of power load based on GM-LSSVM model.First,the GM model is improved by the method of residual correction and background value correction,the residual sequence is obtained by using the improved GM model,and then the prediction is made by LSSVM model.Kernel Principal Component Analysis algorithm(KPCA)can process nonlinear data.The original data can be mapped to high latitude space by nonlinear mapping,so that the data can be preprocessed and the feature information can be extracted.In view of the problem that the single kernel function can not take into account both learning ability and generalization ability.In this paper,we propose a kernel function which is a mixture of polynomial kernel and Gaussian radial basis kernel,that is,MKPCA.RBFNN neural network has excellent performance in generalization ability,clustering analysis ability,information processing ability,multi-dimensional nonlinearmapping ability,global optimal approximation ability,etc.Based on these two models,this paper proposes a combined forecasting model based on MKPCA-RBFNN.Firstly,MKPCA algorithm extracts features and reduces dimensions of power load related data,and then uses the extracted principal components as the input of RBFNN neural network to get the power load prediction results,last,the result of power load forecasting is obtained,so as to improve the accuracy of power load forecasting.According to the characteristics of power load data nonlinearity,uncertainty and complexity,combined with the advantages of GM-LSSVM model,MKPCA algorithm and RBFNN,this paper proposes a combined forecasting model based on GM-LSSVM-MKPCA-RBFNN.First of all,the GM-LSSVM model is used to predict the sample correction sequence;Then,the MKPCA algorithm is used to reduce the dimension of correction sequence,daily maximum temperature,daily minimum temperature,week type and other daily load factors;Finally,several main components with more than 90% cumulative contribution rate are used as input layer neurons to input RBFNN for training,and the power load forecasting results are obtained.Through the MATLAB simulation experiment,and compared with the other two combined forecasting models,the results show that the combined forecasting model can effectively improve the accuracy of power load forecasting,and reasonably analyze and predict the actual power grid data.