DiffCharge: Generating EV Charging Scenarios via a Denoising Diffusion Model

Siyang Li, Hui Xiong, Yize Chen

Abstract: Recent proliferation of electric vehicle (EV) charging load has imposed vital stress on power grid. The stochasticity and volatility of EV charging behaviors render it challenging to manipulate the uncertain charging demand for grid operations and charging management. Charging scenario generation can serve for future EV integration by modeling charging load uncertainties and simulating various realistic charging sessions. To this end, we propose a denoising Diffusion-based Charging scenario generation model coined DiffCharge, which is capable of yielding both battery-level and station-level EV charging time-series data with distinct temporal properties. In principle, the devised model can progressively convert the simply known Gaussian noise to genuine charging demand profiles by learning a parameterized reversal of the forward diffusion process. Besides, we leverage the multi-head self-attention mechanism and prior conditions to capture the unique temporal correlations associated with battery or charging station types in actual charging dynamics. Moreover, we validate the superior generative capacity of DiffCharge on a real-world dataset involving ample charging session records, and attest the efficacy of produced charging scenarios on a practical EV operation problem in the day-ahead electricity market.