Research On Ultra-short-term Wind Power Generation Prediction Based On Deep Learning

Our country is in a critical period of transforming development mode,optimizing economic structure,and transforming growth momentum.In terms of energy,it is being gradually transformed from fossil energy drive to light energy,wind energy and electric energy.As a kind of clean energy,wind energy plays an indispensable role in optimizing the structure of production capacity.It has played a huge role in promoting the construction of a resource-saving and environment-friendly society.However,due to the volatility and intermittentness of wind in the utilization of wind energy,the overall power dispatch of the grid company has been severely tested.Therefore,the ability to make accurate predictions is particularly critical.In recent years,due to the rapid development of artificial intelligence and the great improvement of computing power,it has become possible to dig out the characteristics of data from historical data,and then use historical data to accurately predict wind power.This paper mainly uses deep learning models such as recurrent neural networks,long and short-term memory networks,BP neural networks,gated recurrent units,as well as support vector regression machines and nearest neighbor methods for ultra-short-term wind power prediction.And explore the performance of various methods on the actual data set of a wind farm in Inner Mongolia,analyze the inadequacy of the above methods in the experimental data set,and then demonstrate the suitability of the gated loop unit,and because of the deep learning parameter adjustment work The complexity of PSO-GRU proposed a PSO-GRU model,using particle swarm optimization algorithm to optimize the hyperparameters of the gated loop unit for ultra-short-term wind power prediction.The results show that this method can improve the prediction accuracy of the gated loop unit,and has higher prediction accuracy than the above several independent prediction models,and performs well in actual predictions.The main research contents are as follows:First of all,a brief introduction to the background and significance of the topic selection of this topic is made,and the analysis shows that the wind power industry is developing vigorously.Accurate wind power prediction is of great significance to power grid dispatching and control.Next,the current research status at home and abroad is introduced,the main methods and model combinations currently used for wind power prediction are introduced,and the analysis draws the conclusion that the direction of ultra-short-term wind power prediction requires high timeliness.Relatively few papers have been published in the field of wind power prediction,and research is slightly lacking.Next,the data set used in this experiment is introduced,the reasons for the missing values of the experimental data set are analyzed,and the methods used for the experimental data preprocessing are data normalization and median filling,etc.,and the different characteristics are discussed.The experimental results have different effects.Finally,the ultra-short-term wind power prediction is carried out,and a variety of neural network methods and regression methods are used as comparative experiments.After analyzing the shortcomings of various neural network methods and the directions that can be improved,the PSO-GRU model is proposed to verify the proposed model.Accuracy and practicality.