Electricity Price Forecast Based On Entropy Theory In Electricity Market

With the deepening of the electric power reform,the competition in the power generation industry is becoming more and more fierce,and the electricity industry has become liberalized and marketed,and electricity can trade freely in the market environment like goods.To a large extent,electricity prices reflect the relationship between supply and demand in the electricity market.Therefore,the theory of electricity price has become an important research object in current electric power science.Compared with the load,electricity price has very strong volatility,which is mainly influenced by uncertain factors such as market member behavior,transmission congestion,generator pricing model and so on.These factors can not be incorporated into the forecasting model of electricity price,and these factors can't be incorporated into the forecasting model of electricity price.Make the forecasting of electricity price more difficult than the forecasting of load Therefore,it is very meaningful to use the historical data of electricity market to forecast the future electricity price.This paper describes the characteristics,contents and methods of electricity price forecasting in power market,analyzes and compares various methods of short-term price forecasting.This paper proposes a combined forecasting model,which combines three forecasting models: extreme learning machine model(ELM),neural Network Model(BP)and improved support vector machine model(CS-SVM),based on the theory of information entropy.The main work is as follows:In order to verify the applicability of the proposed model,the paper uses the data of the next spring,summer,autumn and winter as an example to forecast the electricity market in advance for half an hour.Three kinds of single prediction models are tested and their prediction errors are obtained by experiments.Because of the advantages and disadvantages of all single models,the prediction results are different in different cases.According to the error variability of each model,the weight coefficient of each single model in the combination model is determined by using the information entropy theory.Because the ELM model,BP model and SVM model do not have the function of data stabilization,and the fluctuation of electricity price is very large,it is necessary to usesignal processing technology to pre-process the electricity price before forecasting the electricity price.For the ELM model,this paper proposes a two-level decomposition technique combined with ELM,and sets empirical mode decomposition decomposes the original price series into several intrinsic mode functions(IMFs),in order to reduce the instability of data series.The variational mode decomposition further decomposes the high frequency IMFs into multiple modes,and then sets up the ELM prediction model for all the sequences.For the BP model,In this paper,empirical wavelet transform is used to decompose the original sequence into several modal components,and the BP prediction model is established for all the components.For the SVM model,the trend and oscillation components of the price series are extracted by singular spectrum analysis(SSA),the quasi-periodic signal components are reconstructed,and the reconstructed series are modeled and predicted by SVM.In order to improve the prediction accuracy,the cuckoo search algorithm(CS)is proposed to optimize the parameters of the SVM prediction model,and all the prediction sequences are superimposed at the end of the paper,aiming at the problem of over-dependence on the kernel function in the prediction model,and its parameters easily fall into local optimization,the final forecast of electricity price is obtained.Finally,the simulation results show that the combined forecasting model surpasses the optimal single forecasting model in different seasons,and effectively improves the accuracy of electricity price forecasting.