A data-driven multi-objective optimization strategy for coordinated EV charging based on user review mining

To address the issues of grid load fluctuation and reduced user satisfaction caused by uncontrolled charging, this paper proposes a data-model dual-driven coordinated charging optimization framework that integrates demand perception, optimization solving, and decision evaluation. The BERTopic topic modeling is employed to extract “queuing time” and “charging cost” from user reviews as optimization objectives, which are combined with the grid-side load standard deviation to formulate a multi-objective model. The MOGNDO algorithm is enhanced with quasi-opposition-based learning and an elite strategy, and its performance is compared against NSGA-II and NSGA-III. Finally, the entropy weight method combined with CODAS is applied for decision-making. The results show that the improved MOGNDO outperforms NSGA-II and NSGA-III in both convergence speed and dominance relationships, with dominance proportions of 0.269 ± 0.044 and 0.353 ± 0.075 over the latter two algorithms, respectively. The comprehensive optimal solutions selected by all three algorithms outperform uncontrolled charging across all three objectives. This study provides flexible decision support for charging operators in balancing grid friendliness, user experience, and cost control.