Short-Term Prediction Of Wind Power Based On Deep Learning

With the large-scale construction of wind farms,accurate wind power prediction has become one of the key technologies for efficient wind power consumption.In recent years,China's renewable energy-related policies have been reformed and gradually improved,making wind energy a clean energy source for development.Therefore,its proportion in overall power generation has gradually increased,wind power has the characteristics of volatility,intermittentness and uncertainty,when the wind power grid-connected power exceeds a certain range and the power generation cannot be accurately controlled,it will adversely affect the stable operation of the power grid.The related technologies of forecasting and control for wind power generation need to be improved.Firstly,the principle of wind-generated electricity is expounded,and the influencing factors in the process of wind power generation are analyzed.Since the deep learning process takes a long time,in order to save the training time in the prediction process,influencing factors need to be screened.Under the condition that many influencing factors of wind power coexist,the conditional mutual information is used to extract the characteristic variables which have great influence on wind power as the input of the prediction model.The selected characteristic variables are used as prediction input data,and then the model is trained to obtain the final prediction model.Therefore,the accuracy of the data has a certain influence on the prediction accuracy of the model.Secondly,This paper proposes to identify wind power anomaly data based on improved isolated forests.This method can effectively identify wind power anomaly data,which has great significance for improving the prediction accuracy of the model.Finally,based on the randomness and uncertainty characteristics of wind power time series,a method of wind power prediction based on Adaptive LSTM(Adaptive Long Short-Term Memory)is proposed.In order to overcome the shortcomings of subjective setting of hyperparameters in LSTM predictive model training,it is proposed to optimize the hyperparaxneters of predictive models by using GA(genetic algorithm).At the same time,in order to avoid falling into the hyper-parameter local optimum,the genetic algorithm is improved to obtain the optimal prediction model.By comparing the example data with the prediction results of various prediction methods,it is verified that the proposed adaptive LSTM prediction model is not only effective but also has high wind power prediction accuracy.