Research On Wind Power Combination Prediction Method Based On Dual Attention Mechanism And Error Correction

Due to the exhaustion of fossil energy and the environmental problems it causes,the application of renewable energy such as wind energy has been widely paid attention to,and generating wind power has become an important component of our country’s energy system.However,the intermittency and volatility of wind energy affect the consumption of wind power,and the increasing proportion of wind power may also have a negative impact on the smooth operation of the power system.Therefore,accurate wind power prediction is of great significance to the safe and stable operation of the power grid and the realization of the national "double carbon target".Therefore,this paper carries out research on the above situation,and the main work includes:(1)In order to solve the problem of a large number of outliers and missing values in the measured wind power data of wind farms due to various reasons,this paper adopts the quartile method,cubic spline interpolation method and Pearson correlation coefficient to preprocess the data set,which is helpful to establish a more accurate prediction model.Firstly,the outliers in the data set were removed by the quartile method,and the eliminated values and original missing values were filled by cubic spline interpolation,which improved the adverse impact of abnormal data on the subsequent research.Then Pearson correlation coefficient is used to analyze the characteristic correlation.The results show that wind speed,wind direction and temperature have relatively high correlation with wind power,and other features with low correlation are eliminated and finally normalized.(2)In the current wind power point prediction model,the influence of meteorological characteristics on wind power is usually not considered,and features are extracted with the same weight,and fixed weight is adopted in feature extraction,ignoring the difference of influence of different meteorological characteristics and characteristics at different times on wind power.This paper combines the Attention Mechanism(AM),convolutional neural network(CNN),Bi-directional Long Short-Term Memory(Bi LSTM)and Light Gradient Boosting Machine(Light GBM),an ultra-short term point prediction model(CNN-Bi LSTM-AM-Light GBM)based on dual attention mechanism CNN-Bi LSTM and Light GBM error correction is proposed.By integrating the attention mechanism into the two stages of CNN network feature extraction and Bi LSTM network timing prediction,the feature attention module and time attention module are formed,and different weights are given to the data of different features and different moments adaptively,so as to improve the prediction accuracy of the model.Then using Light GBM algorithm to correct the error of the preliminary prediction results to get the final predicted value of wind power point.Experimental comparison is conducted with actual measured data of a wind farm in the northwest of our country.The result has proved the excellent performance of the CNN-Bi LSTM-AM-Light GBM combination model in wind power prediction,improving the accuracy of the ultra-short term prediction of wind power.(3)In order to more effectively reflect the intermittency and volatility of wind power and provide more information for wind power grid-connection and scheduling,a Quantile regression(QR)and CNN-Bi LSTM-AM interval prediction model was proposed.According to the prediction results of different loci,The prediction intervals of the model at 80% and 90% confidence levels were established,and then the probability density curve was obtained by Gaussian kernel density estimation method.The experimental comparison shows that the proposed model can obtain more reliable prediction results and narrow prediction intervals under the same confidence level.