Research On Hybrid Model Of Short-Term Wind Speed Prediction Based On Quadratic Decomposition

The rapid economic development has led to the continuous increase in the consumption of non-renewable resources,and the harm caused to the energy itself and the external environment has attracted widespread attention.The development of clean and renewable energy to reduce the original energy utilization rate has become the core essence of environmental protection,energy conservation and emission reduction,and the realization of high-quality and sustainable development of human society,and wind energy has quickly become the focus of the energy industry due to its high economic efficiency and clean and renewable characteristics.However,wind power generation is easily affected by a variety of factors,resulting in nonlinear and random fluctuations.If large-scale wind turbines are directly integrated into the power grid,it will have a great impact on the dispatch and stability of the power system.A large number of studies have shown that efficient wind speed prediction is an important way to solve the above problems of wind power generation.Therefore,this paper proposes a CEEMDAN-CSVMD-CS-LSTM-ELM-SVR hybrid model with a comprehensive decomposition,optimization and prediction method for the stochastic volatility characteristics of short-term wind speed sequence,so as to carry out research on short-term wind speed prediction.First of all,this paper combines two decomposition algorithms to decompose the original wind speed sequence twice,and uses CEEMDAN to decompose the original sequence once to extract different feature information of the sequence to reduce the non-stationarity of the sequence,and then calculate the sample entropy value of each component,the components whose entropy value is higher than the original sequence sample entropy value are decomposed twice by the VMD（CSVMD）optimized by the CS algorithm,and the different feature information contained in the complex components is further extracted to reduce the complexity of the components.Secondly,for all the subsequences generated by the two decompositions,the prediction model is determined according to the sample entropy value,and the prediction model is composed of SVR,ELM and LSTM optimized by CS algorithm,and the high-degree components are predicted by the CS-LSTM model,the low-complexity components are predicted by the CS-SVR model,and the other components are predicted by the CS-ELM model.Finally,the corresponding prediction values of different components are added to obtain the original sequence prediction result.This paper selects the hourly wind speed data of four meteorological stations in Shandong Province,and sets up three sets of comparative experiments.After verification,the decomposition,optimization and prediction methods in the model proposed in this paper have shown excellent performance.Through the secondary decomposition,the effective information of different characteristics of the sequence is fully extracted,and the non-stationarity of the sequence is reduced.By optimizing the cuckoo search algorithm used in the model,the sequence prediction accuracy is significantly improved.The combination of different forecasting methods improves the overall forecasting accuracy of the model,shows good forecasting effects for four sequences with different volatility,and improves the robustness of the model.