Optimization of plug-in electric vehicle charging with forecasted price

This paper proposes a new method for scheduling the charging of plug-in electric vehicle's (PEV) battery. The method is employed in the demand side management of smart grids and has the goal of reducing the cost of charging over a long time horizon. The problem of scheduling the PEV battery charging is modeled as a Markov decision process with unknown transition probabilities. A fitted Qiteration batch reinforcement learning algorithm with kernel-based approximation of the value iteration is proposed for learning the transition dynamics and solving the charging problem. The solution is obtained based on the knowledge of the true day-ahead electricity prices and predicted prices for the second day ahead. Simulation results using true pricing data demonstrate cost savings of 8%-40% for the consumer.