Short-term Wind Power Interval Prediction Based On Deep Learning

The power generated by wind power is fluctuating and random,so large-scale wind power grid integration will bring many challenges to the power system.To address the uncertainty in wind power generation,it is necessary to conduct research on uncertainty interval prediction.This can help mitigate the adverse effects caused by the high volatility of wind power on the electrical system.More accurate interval prediction results can provide valuable information,aiding in the guidance of grid scheduling and the integration of large-scale wind power generation.Therefore,this study is of great significance for solving the problems faced by large-scale wind power grid integration.A weighted combination of the predictions from the two models.The main research work is as follows:Using the abnormal data detection and cleaning method based on the quartile method and the missing data filling method based on the K adjacent value algorithm to accurately obtain the characteristics of wind power data,by analyzing the correlation coefficient between the characteristics of wind power data and wind power,based on the method proposed in this paper After data preprocessing,the correlation coefficients of wind speed and wind direction to actual power have increased by 4.3%and 1.4%respectively.The results show that the proposed method model can effectively improve the correlation between wind power data and wind power.A dynamic Bayesian network model and a multi-algorithm fusion model were built to predict the two evaluation indicators of interval coverage and interval average width.The comparison and analysis of the two models showed that the interval coverage rate In the indicator,the output result of the dynamic Bayesian network model is 2.46%higher than the output result of the multialgorithm fusion model on average;in the interval coverage width index,the output result of the multi-algorithm fusion model is narrower than the output result of the dynamic Bayesian network model by an average of 13.32%.A weighted interval combination method based on entropy weighting is proposed to leverage the predictive advantages of two models and enhance the quality of the prediction intervals.This method calculates the information entropy weights of the two models separately,and combines the upper and lower boundaries of their predictions with variable weight intervals.The experimental results show that this method reduces the average interval width by 28.21%compared with the dynamic Bayesian model,and increases the interval coverage by 2.46%compared with the multi-algorithm fusion model.Therefore,this method can be used to improve the accuracy and confidence of prediction intervals.The final comparative analysis of the prediction results for low fluctuation range,medium fluctuation range,and high fluctuation range of power indicates that the prediction method based on deep learning and weighted combination can improve the interval coverage rate and narrow the width of the intervals as a whole.This validates the superiority of the proposed model.