Electric vehicle charging load forecasting based on CNN-GAN and semi-supervised regression

Abstract

With the increasing proportion of electric vehicle users in transportation users， their charging behavior dramatically influences the power system operation. Therefore， it is crucial to predict the charging load of electric vehicles accurately. In this regard， a charging load prediction method is proposed based on CNN-GAN （convolutional neural network-generative adversarial network） and semi-supervised regression. A GMM （Gaussian mixture model） is used for cluster analysis of the user samples and extraction of the typical user behavior features. Given the influence of historical data and weather information such as rainfall and temperature， the EV load prediction model groups based on CNN-GAN are built， and the prediction results are obtained by semi-supervised regression. The EV data from a region of East China are used to compare the prediction results and evaluation indexes of several methods. The results show that the prediction model based on CNN-GAN is superior to other methods in prediction accuracy， and the feasibility of the proposed method is verified.

Keywords

• cnn-gan
• semi-supervised regression
• electric vehicle
• charging load prediction