Research On Power Load Forecasting Of Smart Grid Based On SVM And Intelligent Algorithm

A key foundation for ensuring efficient,stable and intelligent operation of power system is the accurate prediction of power load.As a great forecasting tool,support vector machine(SVM)is widely used in power load forecasting.In the smart grid environment,the complexity of users' power consumption pattern and power load fluctuation are increased,and the prediction results of the traditional and single load forecasting models are no longer accurate.In this dissertation,real-time price is introduced as the feature of power load,and the traditional SVM prediction model is improved from data decomposition,feature selection and parameter optimization.Firstly,this dissertation reviews the existing research and related factors in the field of power load forecasting,and introduces the theoretical background.Secondly,the real-time price is introduced as a feature of power load,and the historical load sequence of the holiday to be predicted is selected by the weighted gray relation projection(WGRP)algorithm,while the particle swarm optimization algorithm with second-order oscillation and repulsive force factor(Sec RPSO)is applied to optimize parameters of the support vector machine.Then,the feature selection method is added to the forecasting model,features are selected by the minimal redundancy maximal relevance(m RMR).Finally,empirical mode decomposition(EMD)is introduced into the forecasting mode,and a new hybrid power load forecasting model which considers data decomposition,feature selection and parameter optimization is proposed.Based on the data of Singapore power market,the proposed forecasting models are simulated,and their superiorities in data decomposition,feature selection and parameter optimization are verified by comparing experiments.The simulations show that:(1)Introducing the real time price as a feature of power load and selecting the historical load sequence of the holiday to be predicted via the WGRP algorithm,not only make the feature selection more reasonable,but also improve the prediction accuracy.(2)The Sec RPSO algorithm is superior to genetic algorithm in parameter optimization.(3)Features selected by m RMR contain rich information and the redundancy between each feature is minimal.(4)By using EMD to decompose the original load data,the complexity of operation is reduced while the prediction accuracy is improved.It can be concluded that works done in this dissertation in terms of data decomposition,feature selection and parameter optimization are effective,and the forecasting models established in this dissertation are accurate and feasible.