Multi-objective optimization method for charging and discharging of electric vehicles via Q-learning-based grey wolf algorithm

With the widespread adoption of electric vehicles (EVs), their large-scale integration into the power grid may lead to uncoordinated charging issues, resulting in a significant increase in peak power loads. This may adversely affect the safe, stable operation and economic efficiency of the power grid. To address these challenges, this paper proposed a multi-objective optimization method for EV charging and discharging via Q-learning-based grey wolf optimizer (GWO). The proposed method follows a systematic approach: First, the system architecture was established, defining the roles of EVs and flexible devices in the charging and discharging scheduling process. Second, a multi-objective optimization dispatch model was formulated, with the objectives of minimizing system operation cost, load mean square error, and load peak-to-valley difference. The model was subject to constraints related to power balance and the operational limits of various devices. Third, an advanced multi-objective grey wolf optimizer with integrated Q-learning was developed to address the dispatch model. This optimizer aimed to achieve the Pareto front. Parameters were optimized via orthogonal experiments. The optimizer's effectiveness was measured using metrics assessing distribution and convergence. To address the problem, a case study was used to assess the performance of the proposed Q-GWO algorithm compared to benchmark algorithms, analyzing optimization outcomes and algorithm strengths. Case studies demonstrated 17.3% cost reduction and 31.5% lower peak-valley difference versus benchmark algorithms. Q-GWO's convergence speed (validated via IGD/HV metrics) outperformed GWO by 15.2%, proving efficacy for real-world V2G dispatch. Additionally, the TOPSIS multi-criteria decision analysis approach was employed to identify an optimal solution.