Research On Wind Power Prediction Based On Data Mining And Deep Learnin

Compared with nuclear energy,tidal energy,geothermal energy,wind energy has the advantages of large storage,renewable,easy access,safety and reliability.Over the past decade,wind power installations have been growing globally,and China has gradually become a major wind power installer country.Improving the accuracy of short-term wind power prediction is not only helpful to solve the problem of wind abandonment,but also can realize the economic dispatch of power grid and improve the safety and reliability of operation.In this paper,a wind power point prediction model based on Orthogonal Maximal Information Coefficient(OMIC)-deep learning-Dynamic Mode Decomposition(DMD)is proposed,which considers the influence of wind turbine’s features on wind power output and takes full advantage of prediction errors.In addition,to quantify the uncertainty of wind power output.This paper presents an Orthogonal Maximal Information Coefficient(OMIC)-Bidirectional Long Short-Term Memory(Bi LSTM)-Quantile Regression(QR)interval prediction model.The influence of wind speed,temperature,humidity,wind direction on wind power output is analyzed comprehensively.A feature selection method based on OMIC and deep learning model is proposed,and the feature dimension adapted to the prediction model can be selected.In view of the inherent errors that will occur in the training of prediction models,DMD is proposed to track the spatial and temporal modes of error data to achieve faster and easier error prediction.The prediction model is optimized by feature selection and error correction to obtain more accurate point prediction results.In order to achieve good interval prediction results,OMIC-Bi LSTM is used to select the appropriate feature dimension as the input of prediction model,Bi LSTM’s learning advantage in time series data prediction is used,and a wind power interval prediction model based on OMIC-Bi LSTM-QR is established by combining quantile regression.Numerical examples are used to simulate the actual data of the wind turbine and wind farm,and the prediction index of the proposed model is compared with that of various models.The simulation results show that the wind power point prediction based on OMIC-deep learning-DMD presented in this paper is closer to the real data.OMIC-Bi LSTM-QR wind power interval prediction model can guarantee higher interval coverage with the narrowest average width.