| As one of the most representative clean energy sources in China,wind energy is not only rich in resources,but also mature in development and utilization technology,which has great potential for commercialization.However,with the accelerated increase of penetration rate of wind power in the power grid,the nonlinearity and volatility of wind speed,as a direct factor affecting the output of wind power,will directly affect the dispatching and safe operation of the power grid system,and reduce the power quality.In the current research,accurate and stable wind speed prediction is one of the important measures to improve the reliability of wind power generation.Therefore,this paper comprehensively studies the multi-step prediction model of wind speed time series by combining three aspects: the preprocessing of original wind speed data,intelligent optimization algorithm for hyperparameters and machine learning regression algorithm.The main contents of this paper are as follows:(1)In order to solve the problems of prediction difficulty and low accuracy caused by noise and unstable characteristics of the original wind speed time series,it is necessary to preprocess the original wind speed data.Firstly,the adaptive wavelet threshold denoising(AWTD)is used to improve the quality and validity of data.The evaluation index of denoising effect shows that this method can remove high-frequency noise while retaining more effective information.Then,this paper proposed a two-stage mode decomposition strategy of ESMD-FO-BSO-VMD,which combines extreme-point symmetric mode decomposition(ESMD)with variational mode decomposition(VMD)of parameters ? and K optimized by fractional-order beetle swarm optimization algorithm(FO-BSO).By introducing the diversity entropy,the high entropy component of the ESMD primary mode decomposition is further decomposed by FO-BSO-VMD.The decomposition results show that there is no mode aliasing in each modal component,and the complexity of high entropy components is further weakened,which proves the effectiveness of the two-stage mode decomposition strategy in dealing with complex time series,and it can provide high-quality data sets that are more easily used for prediction experiments.(2)The machine learning algorithms involved in the wind speed prediction model in the current research mostly have the requirement of hyperparametric optimization.Aiming at the problems that the original beetle swarm optimization algorithm(BSO)is easy to fall into the local optimal value and has slow convergence speed,this paper proposed an improved fractional-order beetle swarm optimization algorithm(FO-BSO)based on the memory of the G-L definition of fractional calculus.Through the benchmark function test experiment,the parameter sensitivity analysis of derivative order β and memory term r,the algorithm execution time analysis and the comparative analysis of various optimization algorithms are carried out.The experimental results show that the FO-BSO(β(28)0.1,r(28)4)algorithm has superior performance in accuracy,stability and convergence speed.And this optimization algorithm is used to intelligently optimize the relevant parameters of VMD algorithm and machine learning prediction algorithm in this paper.(3)Based on the two-stage mode decomposition,FO-BSO algorithm,the complementarity of recursive multi-step prediction and direct multi-step prediction,and the combined prediction of modal components with different complexity by long short term memory network(LSTM),least squares support vector machine(LSSVM)of machine-learning and auto-regression integrated and moving average(ARIMA)of statistical method,a hybrid model of multi-step wind speed prediction is constructed.The comparative experiments of multi-step prediction with different mode decomposition strategies show that the primary decomposition ESMD can significantly improve the accuracy of the prediction model,and the two-stage decomposition strategy of ESMD-FO-BSO-VMD can further improve the performance of the model on the basis of the first decomposition.The data processing mode of two-stage mode decomposition can effectively improve the quality of data sets and enhance the multi-step prediction effect of the model.The comparative experiment results of multi-step prediction of different prediction models show that the FO-BSO algorithm and the combined prediction methods of LSTM,LSSVM and ARIMA proposed in this paper can effectively improve the accuracy and stability of the hybrid prediction model.And the complementary advantages of direct multi-step prediction and recursive multi-step prediction can alleviate the problem of rapid decline of model performance caused by the increase of prediction step.In addition,the simulation results of three groups of test data with different sampling locations and sampling frequencies further illustrate that the ESMD-FO-BSO-VMD-LSTM-LSSVM-ARIMA hybrid prediction model constructed in this paper not only has excellent multi-step wind speed prediction performance,but also has strong generalization ability. |
