A Forecasting Method For Wind Power Based On Improved Long-short Term Memory Neural Network

With the reduction of fossil energy year by year and the proposal of the national dual-carbon strategy,new energy is connected to the grid on a large scale,forming a new power system with new energy as the main body.As the penetration rate of new energy continues to increase,the volatility and randomness of new energy itself will affect the stability of the power system and cannot fully consume new energy.Accurate wind power prediction helps the power system dispatch department to obtain the pre-output plan of the wind farm,and improve the new energy consumption capacity of the power system by coordinating various power generation resources.This thesis mainly focuses on the research on wind power characteristics and wind power prediction.The key tasks are summarized as follows:（1）The basic theoretical knowledge related to wind forecasting is sorted out.Firstly,the influencing factors of wind power generation are introduced,including wind speed,height,wind direction,temperature and other factors.Secondly,several commonly used data processing methods are introduced,which are Pearson correlation coefficient method and principal component analysis method.Finally,the evaluation methods of the prediction results are expounded,including mean absolute error,mean absolute percentage error and mean square error respectively.（2）Aiming at the randomness and volatility of the wind power sequence itself and the long dependence on the time length,a wind power prediction method based on an improved long short-term memory neural network is proposed.Considering that there are many influencing factors of wind power,the Pearson correlation coefficient method and principal component analysis method are used to select the series of influencing factors according to the degree of correlation,and the influencing factors with high degree of correlation are selected as the input of the prediction model;Long-term dependence,long-term and short-term memory neural network is used to build a prediction model,and for the randomization of key parameters of the prediction model,the cuckoo algorithm is used to optimize the key parameters to improve the accuracy of the prediction model.Using the wind farm data in a certain area of Gansu to conduct simulation analysis,the results show that the long-term and short-term memory neural network can effectively predict the wind power.（3）Considering the limitations of different forecasting methods,an LSTM hybrid wind power forecasting model based on error correction is proposed.The non-recursive variational modal decomposition is used to effectively decompose the wind power sequence,and multiple modal components are obtained.Considering the possible modal aliasing phenomenon of modal components,the sample entropy algorithm is used to analyze the decomposed modal components.Reconstruction;using long short-term memory neural network to build the prediction model of each modal component,and introducing the firefly optimization algorithm to optimize the key parameters of the prediction model to improve the overall convergence speed of the model;considering the limitations of the prediction model itself,from the perspective of prediction error Starting from the beginning,the least squares support vector machine method is used to perform secondary prediction on the prediction error sequence,so as to correct the wind power prediction result of the prediction model.The simulation results show that the hybrid prediction model can avoid the limitations of a single algorithm and effectively improve the accuracy of the prediction results;error correction can improve the prediction accuracy problem caused by the prediction model architecture,and further improve the effectiveness and accuracy of the prediction model.