Research On Short-Term Wind Speed Prediction Based On ARIMA-BP-PSOLSSVM Combined Optimization Model

With the raised requirements of ecological development in China,energy conservation and emission reduction has become the top priority of the whole society in recent years and the most effective way is vigorously develop renewable energy.As a kind of irreplaceable green renewable energy,wind energy has the characteristics of easy development and wide sources,but due to the climate and geographical location reasons it also has strong volatility and randomness,which is the biggest obstacle to the wide development.With the development of the power market,the prediction technology of wind speed could improve the stability of wind energy into the power grid which make this renewable energy more competitive.The short-term wind speed prediction model proposed in this paper named as the VMD-PSR-(ARIMA,BP,PSOLSSVM)-PSOLSSVM which refer to as ARIMA-BP-PSOLSSVM combinatorial optimization model.The model combines the newly proposed variational mode decomposition,phase space reconstruction and three typical linear and nonlinear prediction methods.Firstly,the proposed combination model uses VMD(Variational Mode Decomposition)to modally decompose the original wind speed data and obtained a series of more stable components that can represent the original data and a residual term.Secondly,the model adopts PSR(Phase Space Reconstruction)to reconstruction the VMD result and the residual term.The PSR processed data can be used as the input and output matrix of each component in the artificial intelligence prediction model.Thirdly,according to the effective combination of models can solve the partially extract information and model selection,this paper selected three typical methods to predicate the short-term wind speed at the same time and combined them to optimize to improve prediction accuracy and stability.The typical linear model is ARIMA(Autoregressive Integrated Moving Average Model)and two classic nonlinear models with different principle are BP(back propagation neural network)and PSOLSSVM(Particle Swarm Optimization Least Squares Support Vector Machine).Finally,the PSOLSSVM method is selected as the nonlinear combination method to optimize the prediction results of three separate prediction models.The specific implementation step is using the prediction results of three separate prediction methods,ARIMA,BP and PSOLSSVM as the input of PSOLSSVM for secondary prediction.As a result,the secondary prediction can make good for deficiency of individual models to achieve the purpose of combinatorial optimization.The proposed combination optimization model is applied in the two fluctuating data sets of Cheng De Paddock wind farm in this paper,and the prediction results are compared with other eight models from error comparison analysis and method improvement analysis.The results of the analysis verify that the proposed model has great advantages that surmount any other decomposition methods,individual prediction models or other combination methods in terms of prediction accuracy and prediction stability.The large-scale development of wind energy is the general trend of the global energy transformation,at the same time,the improvement of short-term wind speed prediction effectiveness is convenient for wind farms to control the number of wind turbines running and operating costs,and to make planning of fan maintenance.It can also help the dispatching department adjust the plan in time to reduce the impact of wind energy on the power grid.The combined optimization model proposed in this study can effectively and stably predict the short-term wind speed,which will have a positive impact on the development of wind energy and the dispatch of the power grid.