The Research And Application Of The Combination Algorithm Based On Artificial Intelligence And The Multi-step Ahead Forecasting

Wind energy is crucial importance to the low carbon energy technologies,it makes the realization of the sustainable energy supplies possible.And to some extent,it constitutes a key part of the smart grid infrastructure.However,the instability and randomness of the wind are of vital obstacles to the large-scale application of wind power.Therefore,the demand is higher and higher for the accuracy and performance of the wind speed prediction model.Now,a single model already can't meet the existing requirements,more and more people begin to put attention in the combination forecast model of wind speed based on multi-step ahead forecasting.There are two important problems have to be solved,one is how to choose the suitable model according to the merits of different models to form a combination model and another is how to analysis the influence of sub model on combination model.In this paper,a new combined forecasting model(SMSGECSEP)based on Singular Spectrum Analysis,Sliding Window Model,Multi-step ahead forecasting,General Regression Neural Network,Elman Network,Cascade BP,Echo State Network,Simulated Annealing and Particle Swarm Optimization have been put forward.In the short term wind speed forecasting,forecasting accuracy is hardly affected by the noise signals,which are caused by a lot of erratic factors.Therefore,in the first place,the Singular Spectrum Analysis method is used in SMSGECSEP model to reduce the influence of noise signals.Secondly,the input vector set and output vector set of multistep ahead forecasting are obtained by decompose the raw wind speed data and the de-noised data,respectively.And in this process,the sliding window model is used.Thirdly,using three sub models,GRNN optimized by SA(SA_GRNN),Elman optimized by SA(SA_Elman)and CBP optimized by SA(SA_CBP),models the input vector set and the output vector set.And then ESN which is optimized by PSO(PSO_ESN)is used to combine the above three prediction results.In the process of the above,the parameters of above four neural network are optimized by SA and PSO for the arm of increasing the prediction accuracy.In this paper,there are a total of two series of simulation.The one is that through the simulation of the average wind speed date of the national wind technology center M2 tower,and make comparison the combined model with other three sub models,the following result can be got: plays a positive role in promoting short term wind speed forecasting precision.The another is that through the simulation of other three data series with different levels noise reduction to study the noise reduction performance of SSA,and reached the conclusion that when to get rid of 50% information from wind speed data,the prediction performance of forecasting model is best.