Optimizing Electric Vehicle Charging Scheduling With Deep Q Networks and Long Short‐Term Memory‐Based Electricity and Battery State of Charge Prediction

ABSTRACT The number of electric vehicles (EVs) on the road is rising as a result of recent advancements in EV technology, and EVs are important to the smart grid economy. Demand response schemes involving electric vehicles have the potential to dramatically reduce the cost of charging if real‐time price signals are properly exploited. However, because of unpredictable traffic patterns, fluctuating client commuting patterns, and utility pricing strategies, selecting the optimal charge strategy can be difficult. Traditional model‐based systems need a forecasting and optimization model for the scheduling process. The scheduling problem is modeled in this work using a Markov decision process (MDP) with an uncertain transition probability. A deep reinforcement learning‐based model‐free method is proposed to address this issue as effectively as possible. Without any prior knowledge about the system model, the transition probability can be adaptively learned. The suggested method makes use of two networks: a Q network is used to estimate the optimal action‐value function, and a representation network is used to extract discriminative characteristics from power prices. The effectiveness of the established method has been demonstrated by multiple investigations.