Smart Online Charging Algorithm for Electric Vehicles via Customized Actor–Critic Learning

With the advances in Internet of Things technology, electric vehicles (EVs) have become easier to schedule daily life, which is reshaping load curve. It important design efficient charging algorithms mitigate negative impact EV on power grid. This paper investigates an scheduling problem reduce cost while shaving peak load, under unknown future information about EVs, such as arrival time, departure and demand. First, we formulate minimize electricity bill fleet study online setting without knowing information. We develop actor-critic learning-based smart algorithm (SCA) against uncertainties behaviors. The SCA learns optimal strategy with continuous actions instead discrete approximation charging. further a more computationally customized learning (CALC) by reducing state dimension thus improving computational efficiency. Finally, simulation results show that our proposed can EVs' expected 24.03%, 21.49%, 13.80%, compared Eagerly Charging Algorithm, Online RL-based Adaptive Energy Management respectively. CALC efficient, its performance close only gap 5.56% cost.