Nonparametric User Behavior Prediction for Distributed EV Charging Scheduling

Abstract

We propose a distributed electric vehicle (EV) charging scheduling to minimize load variance in the distribution grid and reduce EV charging cost. To predict the availability and load demand of the EVs, we use nonparametric diffusion-based kernel density estimator (DKDE) to model the stochasticity of charging load. DKDE is based on smoothing properties of linear diffusion process which is more adaptive to the training dataset and results in optimal bandwidth selection comparing to Gaussian kernel density estimator (GKDE). Then, we formulate the optimal charging problem as a sharing problem which is solved efficiently by alternating direction method of multipliers (ADMM). Using real data numerical simulation, we evaluate DKDE prediction accuracy and verify the EV charging scheduling performance.