Research On Some Difficult Problems In Wind Speed Forecasting Combination Strategy

As a basic element in the atmospheric environment,wind research is of great significance for weather,climate,environmental science,clean energy,and meteorological disasters.However,due to many factors such as temperature,pressure,altitude,topography and latitude,wind has the characteristics of randomness,intermittentness and volatility,making wind one of the most difficult to predict weather forecasting elements.Therefore,the research on wind speed prediction methods plays a key role in the improvement of weather forecast,the study of environmental pollution,the development of wind resources and the prevention and control of gale disasters.The wind speed data is highly volatile and random.Direct prediction using a single prediction model usually causes large errors.At the same time,the single model without parameter optimization still has unstable model parameters in the prediction,and the weight of the combined model is difficult in the training process.The identification of other issues has become a hot spot in wind speed prediction research.Therefore,in view of the shortcomings of traditional methods,this paper proposes research schemes to improve prediction accuracy from three perspectives:wind speed data processing,parameter optimization,and combined forecasting,and proposes three different combined optimization prediction models.Combined wind speed prediction based on EEMD decomposition and multi-model parameter optimization,wind speed combined prediction based on new secondary signal decomposition and parameter optimization,and combined weighted prediction based on MOBSFPA multi-objective optimization.Furthermore,based on this,a new wind speed fusion prediction model is proposed,which is based on CEEMDAN decomposition and CLSFPA optimization.The main results are as follows:1)For the previous model,the problems of data preprocessing and model parameter optimization are neglected,and a novel combinatorial learning model is proposed.In this model,the original wind speed data is first divided into a set of finite signal components by the set empirical mode decomposition.The Drosophila optimization algorithm is used to optimize the model parameters,and each signal is predicted by the parameter-optimized artificial intelligence model.The final predicted value is obtained by rebuilding the refined series.In order to estimate the predictive power of the model,the 15-minute wind speed data of wind farms in China's coastal areas were selected as a case study.The empirical results show that the proposed model is superior to the existing similar prediction models.2)Given that wind speed series often have complex features,wind energy prediction is very difficult.Aiming at this challenge,this paper proposes a prediction architecture based on the new hybrid double decomposition technique(HDT)and flower pollination algorithm(FPA)optimized back propagation(BP)neural network prediction algorithm.The proposed HDT combines the complete set of empirical mode decomposition adaptive noise(CEEMDAN)and empirical wavelet transform(EWT).After a decomposition,the EWT is further used to decompose the high frequency eigenmode function(IMF)generated by CEEMDAN,thereby reducing Predicted sequence complexity.Finally,the improved BPNN of the flower pollination algorithm is applied to predict all the decomposed IMFs and patterns.In order to study the prediction ability of the model,multi-step prediction was carried out using wind speed data of two different wind farms in Shandong,China.The experimental results show that the model outperforms all other models in one-to five-step wind speed prediction,indicating that the model is very suitable for non-stationary multi-step wind speed prediction.3)In the past,the combined forecasting model usually only sets a single optimization target(accuracy or stability)in the weight optimization process.This paper introduces a new combination method based on multi-objective optimization algorithm.For the weight coefficient of the combined model,the deviation-variance framework is set as the objective function of the multi-objective optimization problem,and the combined model converges to the best precision and stability at the same time.At the same time,a pollen propagation algorithm based on bat search algorithm is proposed,which combines Pareto optimal theory to form a new multi-objective algorithm.In addition,data denoising is also included in the data preprocessing stage.In order to evaluate the predictive power of the model,12 wind speed data sets of two wind farms in the eastern coastal areas of China were selected as case studies.The experimental results show that the developed multi-objective combination model is superior to other comparison models in terms of high precision and stability of wind speed prediction.4)Based on the above research,a wind speed fusion prediction method is proposed,which combines model optimization,complete set empirical mode decomposition adaptive noise,improved flower pollination algorithm and chaotic local search(CLSFPA)and no negative constraint theory(NNCT).).Firstly,the original CEEMDAN is used to divide the original wind speed data into a finite set of IMF components.According to the characteristics of different decomposition sequences,five optimal prediction models are selected in the single model library by using the model optimization method,and the combined model model is combined with NNCT.To predict each decomposition signal.At the same time,an improved chaotic local search(CLS)-flower pollination algorithm(FPA)is proposed to determine the optimal weight coefficient of the combined model,and finally the final predicted value is obtained through sequence reconstruction.To assess the predictive power of the proposed combined model,15-min wind speed data from four wind farms in the eastern coastal regions of China were used.