The Application Of Combined Forecasting Model Based On Singular Spectrum Analysis In Power Load Forecasting

Electricity,which has attracted particular attention in recent years,is one of the most important energy sources in people's life.With the gradual development of electric power system,electric load forecasting has been proved to be a very useful tool to make decisions of the power system for planning and operation for all market participants in electricity markets.And accurate electric load forecasting can not only ensure the safe operation of power system and improve the quality of power supply,but also contribute to power companies and consumers make rational plans and maximize their profits.However,due to the electric load suffered the impact of various stochastic factors such as climate,economy,and social change and so on,which make it difficult to store in large quantities like coal,oil and other resources in real life,so,electric load forecasting still remains an enormous problem.Considering that a single prediction model can't be better to reflect fully the complex law and information of electric load data,in order to improve the accuracy of electric load forecasting and make full use of the advantages of every model,on the basis of statistical analysis of a large number of historical load data,a new hybrid prediction model based on singular spectrum analysis is proposed.First of all,the paper introduced the background and significance of the research in detail,and summarized the various factors that affect the load forecasting.Based on the development and current situation of electric load technology at home and abroad,the classification and basic steps of load forecasting are discussed in depth,and the advantages and disadvantages of various methods of load forecasting are analyzed and compared.Secondly,in order to improve the prediction accuracy of power load,the singular spectrum analysis(SSA)was used for original signal de-noising based on the analysis of the factors that affect the combined forecasting model.The experimental results showed that the signal that removes the noise component is more smooth and more flat,which is beneficial to the more in-depth analysis.Finally,based on the basic theory and method of combinatorial forecasting,this paper uses the method of linear weighted averaging to construct a hybrid model,which was made up of three individual forecasting models,namely double-layer BP neural network,extreme learning machine(ELM)and echo state network(ESN)optimized by PSO.The hybrid model combined the characteristics of the double-layer BP network and the echo state network,which training process was more simple.At the same time,it made full use of the advantages of extreme learning machine,which has simple operation,shorter operation time and better generalization capability.Compared with the individual model,the hybrid model had better prediction performance.In addition,the paper set the average absolute percentage error(MAPE)as the objective function of the combined forecasting model,and used the simulated annealing algorithm(SA)to optimize the weight coefficient of the combined model so that the objective function value was the smallest and the prediction accuracy was the highest.Based on the proposed model,the paper carried out the short-term simulation prediction taking the load data from Queensland electric market in Australia.The Simulation results showed that the proposed model was of higher prediction accuracy than the single model and had a wider application prospect.