Research On Intelligent Method Of Short-Term Wind Speed Forecasting

The development of economy makes the quality of people's life becoming higher and higher,but the resulting environmental problems are also getting worse and worse,such as air pollution,marine pollution,forest fires,virus ravages,and increasing scarcity of resources such as oil.Therefore,it becomes emergent to make efficient usage of renewable clean energy resources,i.e.wind power.However,the randomness and instability of wind speed cause difficulty in model to predict wind speed accurately.A combined model based on weights is proposed for wind speed in this paper.We discuss the prediction of wind speed mainly from the following three aspects:(1)Method selection for noise reduction in data;(2)Predictive study of single models;(3)Predictive study of combined models.Because of the instability of original wind speed data,the prediction results are not satisfactory by using the original data before the prediction and data reconstruction is needed.We select five methods in this paper: empirical mode decomposition(EMD),integrated empirical mode decomposition(EEMD),wavelet transform(WD),Gaussian filter and mean filter are separately used to deal with the noise reduction of the original wind speed data at six stations.Doing prediction experiment of the processed data separately based on long and short term memory neural network LSTM,and we find that the prediction effect of EMD and WD methods are obviously better than other methods.Single model predictive selection support several models,i.e.support vector regression SVR,multi-layer perceptron MLP,circulating neural network RNN,long and short term memory neural network LSTM and GRU circulating neural network.Selecting EMD and WD these two methods to reconstruct the original wind speed data.The experiment is divided into two groups,and the results respectively show that the effect of EMD-RNN,EMD-GRU,EMDLSTM and WD-RNN and WD-GRU,WD-RNN,WD-LSTM are better.Selecting four single models of EMD-LSTM,EMD-GRU,WD-NN and WD-GRU as basic elements to construct linear combined model,and selecting the optimization algorithm to determine the weight of the combined model,the combined model of six sites gives a more accurate prediction than a single model at last.