Data-driven spatial-temporal prediction of electric vehicle load profile considering charging behavior

Abstract

Accurately predicting the spatial-temporal distribution of electric vehicles (EVs) load is of great significance to the optimal dispatching and safe operation of the power grid. This paper proposes a spatio-temporal distribution prediction method for EV charging loads, which considers the charging behavior characteristics of different types of EVs and the spatio-temporal coupling between EVs and charging stations. Firstly, an EV charging demand prediction model based on improved random forest (IRF) is established, in which the parameters of random forest (RF) prediction model for each type of EVs are optimized by harmony search (HS) with the principle of minimum prediction error to release the sensitivity of the EV charging prediction model to parameters. Then, a bottom-up method for predicting the spatial and temporal distribution of EV cluster charging loads based on IRF is proposed, which takes into account the spatial-temporal coupling between EVs and charging stations. In addition, a parallel computational method with data parallelization and task parallelization is proposed to enhance the efficiency and practicability of the proposed method. Finally, the accuracy level of IRF has been explored through rigorous case studies comparing with support vector machine (SVM), back propagation neural network (BPNN) and general random forest (RF). The case studies also reveal that the proposed method, compared with traditional methods, can not only improve the prediction accuracy of the total EV charging load but also obtain the spatial and temporal distribution of the charging load in the region.