Research On Short-term Wind Power Intelligent Prediction Technology Based On Combined Model

With the change of the world energy pattern,promoting the development of energy structure to focus on clean energy has become an important goal of China.Wind energy,as a key clean energy,has attracted great attention.However,due to the volatility and randomness of wind energy itself,wind power genera tion will bring great challenges to the stable operation of power grid in the process of grid connection.Therefore,the effective prediction of wind power is of great significance for adjusting the operation of power grid and maintaining the stability of power grid.This paper makes the following research on short-term wind power prediction.Firstly,taking the original data processing of wind farm as the starting point,extract the key feature attributes in the data set,simplify the data set,and use an improved isolated forest algorithm based on sliding window to clean the abnormal power data in the data set,so as to make the processed data set more meet the needs of model operation.Then,based on the obvious characteristics of time series of wind powe r data set,the long-term and short-term memory network LSTM is selected as the prediction method.At the same time,aiming at the problem of high error in the prediction process of single model,the combined prediction model vmd-lstm is improved and designed.The original power data is decomposed by variational modal decomposition VMD to reduce the noise in the original data,obtain the characteristics of the original sequence and improve the prediction accuracy.Finally,aiming at the problems of difficult selection of variational modal decomposition parameters and insufficient prediction accuracy and robustness of the model in vmd-lstm model,an improved bapso-vmd-ln-lstm combined prediction method based on error correction is proposed,and an improved pa rticle swarm optimization algorithm based on longicorn whisker search algorithm is applied to realize the adaptive parameter finding of variational modal decomposition VMD;The long-term and short-term memory network LSTM is optimized by layer normalization algorithm to prevent gradient disappearance or gradient explosion during model training;Aiming at the errors in the prediction process,the prediction error correction strategy is applied to correct the prediction error of the model.After optimization,the robustness and prediction accuracy of the combined model are further improved..Based on the historical data set-up experiment collected by the data system of a wind farm,a variety of prediction models are compared.It is verified that the bapso-vmd-ln-lstm combined model based on error correction proposed in this paper can achieve high accuracy in short-term wind power prediction and can meet the scenes requiring high prediction accuracy.