Research On Deep Learning Fusion Algorithm For Short Term Load Forecasting

Accurate short-term load forecasting is of great significance to the safe,stable and economic operation of power system.In order to achieve the above goals,a basic and important part is to analyze the characteristics of short-term load and the influencing factors of its fluctuation.The traditional short-term load forecasting method is still insufficient in dealing with massive data and nonlinear mapping ability.In recent years,the information technology represented by artificial intelligence technology is making great changes in many industries including finance,medical treatment and security.At the same time,with the continuous deepening of building a strong smart grid in China,China’s power enterprises further put forward the goal of building a ubiquitous electric internet of things,which shows that using a new generation of artificial intelligence technology with deep learning algorithm as the core to solve the problem of short-term load forecasting has a technical basis and industry needs.In order to keep up with the trend of the times,improve the nonlinear mapping ability of the short-term load forecasting method,and explore the new way to solve the short-term load forecasting problem by using the deep learning algorithm,this paper mainly carries out the following research:(1)Based on the actual data,the characteristics of short-term load and the influencing factors of its fluctuation are analyzed,and the data set is constructed.Firstly,the periodic characteristics of short-term load and three main influencing factors of weather,season and date type are determined through qualitative analysis;secondly,the correlation between different weather factors and short-term load is measured by distance correlation coefficient,and the weather factors with strong correlation are screened out;finally,data preprocessing is carried out for the selected factors and historical load data to construct the data set for short-term load forecasting.(2)Aiming at the deterministic prediction of short-term load,a new methodbased on multi neural network fusion is proposed.The core algorithm of this method is parallel architecture,which combines convolutional neural network,gated recurrent unit network,attention mechanism network and maxout network.After training with the above data set,it is used for short-term load deterministic prediction.The prediction results show that this method has advantages.(3)Aiming at the probability prediction of short-term load,a method of short-term load probability prediction based on quantile regression of time convolution network is proposed.This method combines the time convolution network with the quantile regression theory to get the load forecasting results and their probability density distribution under different quantile conditions.The above data sets are used to verify the effectiveness of the method.