Research And Application Of Multiple Hybrid Forecasting Algorithm

In the wind energy industry,how to improve the accuracy of wind speed forecasting has always been a difficult and challenging problem.Accurate wind speed forecasting can guide the mobilization and maintenance of wind turbines,so that it improved the efficiency operation of wind farm and ensured the safety of wind energy integration.However,a single forecasting model or a single hybrid model cannot effectively forecast the different feature of wind speed time series in a wind farm.In view of this,this paper uses the K-means clustering method to gather 50 groups of wind speed time series in a wind farm in Shandong into 3 categories,and randomly selected the 4 groups,the 5 group,6 groups the wind speed time series in different categories to forecast.In the forecasting process,three hybrid models are proposed in this paper(EMD-SDCS-SVM,FEEDM-CGFPA-ABBP and WD-APSOACO-BP).Three hybrid forecasting model of the process is as follows: 1)data preprocessing: using EMD,FEEMD and WD to remove the high frequency noise of three different feature of wind speed time series;2)parameter optimization: in order to improve the performance of single optimization algorithm,this paper proposes three improved optimization algorithm.By using the gradient descent algorithm(SD)to improved cuckoo(CS)algorithm,which improve the convergence speed of the cuckoo algorithm;using conjugate gradient(CG)algorithm to modified the flower pollination algorithm(FPA)to improve the local search ability and convergence speed of pollen dispersal algorithm;adaptive particle swarm optimization algorithm(APSO)to ACO(improved ant colony algorithm),the improved ant colony algorithm to avoid falling into the local optimum,improved ant colony algorithm in convergence speed and reduces the computational complexity of single algorithm.3)forecasting process: using SDCS algorithm to optimize the support vector machine(SVM)the penalty coefficient and kernel function;searching for the optimal parameter k and Bt of the ?-ABBP model;using APSOACO optimization BP neural network the weights of between the input layer and hidden,and the threshold of between the hidden layer and output layer.Numerical results indicate that the three different hybrid model is simple and can satisfactorily approximate the different feature of wind speed series by comparing with individual models and traditional hybrid model.Therefore,the developed hybrid model can be an effective tool in mining and analysis for wind power plants.