Non-myopic optimization of user-centric electric vehicle charging and discharging considering intermittent parking patterns

As Electric Vehicles (EVs) increasingly serve as flexibility resources for power systems, users’ parking patterns have become pivotal in determining the performance of charging and discharging strategies. The intermittency and uncertainty of parking behaviors limit EV availability, undermining the effectiveness of conventional optimization methods that lack foresight across parking epochs. This paper establishes a non-myopic optimization framework for EV charging and discharging, aiming to improve both scheduling revenues and user comfort under intermittent parking patterns. Firstly, a Markov chain-based parking chain generation method is introduced to simulate realistic sequences of intermittent parking epochs and corresponding energy requirements. Secondly, an enhanced Flexibility Feasible Region (eFFR) is formulated to incorporate multi-epoch temporal dependencies into current decisions, with theoretical guarantees on problem solvability and optimality for user-centric scheduling problems. Then, a Deep Reinforcement Learning (DRL) framework is proposed to implement non-myopic optimizations, where tailored state encoding and normalization strategies are designed to better characterize heterogeneous intermittent parking patterns. The results indicate that incorporating intermittent parking patterns can mitigate the range anxiety of EV users. The proposed DRL-based non-myopic optimization method can significantly improve peak-valley arbitrage performance while achieving a notable reduction in the user’s Range Anxiety Rate (RAR) compared to the Mixed-Integer Linear Programming (MILP)-based myopic benchmark and the Model Predictive Control (MPC)-based non-myopic benchmark. Furthermore, ablation studies demonstrate the effectiveness of the proposed DRL method by showing improved policy stability and more consistent decision behavior.