Deep Learning-based Stacking Ensemble for Charging Demand Prediction in Power Grid Using Charging Behavior Pattern of Electric Vehicle Users

Abstract

Nowadays, with the spread of global eco-friendly systems and the development of battery performance, electric vehicles are leading the paradigm shift in the automobile market. However, it can cause a significant problem for the power system when electric vehicle has a high rate of load change over time and multiple users request charging their electric vehicles simultaneously. Therefore, predicting charging demand of electric vehicles is a crucial issue for the stable management of the power network. Previous attempts have been made to predict electric vehicle charging demand using machine learning algorithm and deep learning neural network. However, due to different erratic charging behavior patterns for each vehicle users, it can be a degrading performance by weakening the generalization capacity of the prediction model.

In this paper, we cluster each electric vehicle user’s charging behavior based on three parameters through k-means clustering, unsupervised learning method. We label each user with corresponding clustering class and utilize it as an additional feature of predictive model to solve this problem. In addition, we propose new electric vehicle charging demand prediction algorithm using deep learning based stacking ensemble. The proposed algorithm has been shown to improve the mean absolute error and standard derivation error of electric vehicle charging demand prediction in power grid.