| With the rapid development of national economy,the demand for energy is gradually increasing.The traditional fossil energy which is non renewable becomes more scarce and it also has irreversible harm to the ecological environment.Therefore,with the strong support of the state,renewable clean energy industry is developing rapidly,among which wind power generation has achieved an important development status due to its low cost,strong environmental protection and other advantages.However,due to the intermittence randomness and instability of wind power generation,it brings severe challenges to the safe and stable operation of power grid and power dispatching.Therefore,it is of great significance to realize the accurate prediction of wind speed.Based on the real-time meteorological data collected by a wind farm,this paper proposes two different hybrid wind speed prediction models for different wind speed prediction problems.The main research contents are as follows:Firstly,preprocessing the data.In order to ensure the integrity and accuracy of the experimental data,a small number of abnormal values in the data are directly deleted,and then the vacant values are supplemented by interpolation.Considering the sufficiency and diversity of the experimental data,the sampling period of the original data is adjusted to generate data sets with different sampling periods.The influence of dimensional difference between variables on the experimental results is eliminated by normalization method.Through the above data analysis and processing,this study can be carried out scientifically and effectively.Secondly,aiming at the unstable prediction performance of long-term and short-term memory neural network model,a multi-dimensional long-term and short-term memory neural network prediction model optimized by improved bat algorithm is proposed.By making the initial population more dispersed and evenly distributed in the search interval,the global search ability of bat algorithm is enhanced.At the same time,in order to better mine the law of wind speed change,multiple variables related to wind speed are used as model inputs;the long-term and short-term memory model optimized by the improved bat algorithm is used to predict the wind speed.The experimental results show that the improved bat algorithm can effectively improve the prediction accuracy of the model,and the multi-variable input also makes the model mine the wind speed change characteristics more accurately.Finally,aiming at the problem that it is difficult to predict accurately due to the instability and nonlinear characteristics of wind speed,and considering the practical application conditions of wind speed prediction,a hybrid wind speed prediction model based on real-time double decomposition is proposed.Based on the practical application conditions,two data decomposition methods,discrete wavelet decomposition and variational mode decomposition,are combined to decompose the wind speed data into multiple sub-sequences in real time,which fully and reasonably reduces the nonlinear characteristics of the data,so as to improve the prediction accuracy of the model.At the same time,in order to provide more wind speed information in the future,fuzzy information granulation method is used to predict the maximum and minimum wind speed curves.Experimental results show that the proposed model can effectively improve the generalization ability and prediction performance of the model. |
