<p>Pipeline siltation-induced drainage impairment constitutes one of the primary causes of urban waterlogging. Accurately and quickly identifying silted pipelines and proposing corresponding remediation solutions is of significant importance for urban waterlogging prevention. This study developed a multi-objective optimization model for grey-green coupled drainage systems incorporating trenchless pipeline desilting to mitigate urban waterlogging and reduce construction costs. First, an LSTM-Transformer-based method was developed to diagnose urban drainage pipeline siltation by analyzing flow rate-siltation relationships and integrating hydraulic characteristics with deep learning. Second, a PSO-NSGA-Ⅲ multi-objective optimization method was proposed by enhancing the crossover operator of NSGA-Ⅲ with PSO to improve convergence and computational speed. Finally, the siltation diagnostic results were input into the InfoWorks ICM model to propose a trenchless desilting-based waterlogging mitigation strategy, and its reduction effectiveness was analyzed. The results show that the LSTM-Transformer-based diagnostic model achieves over 85% accuracy in pipeline siltation prediction. The maximum siltation coefficient reaches 0.51, and severe siltation is observed in northeastern pipelines, which is consistent with field survey results. The multi-objective optimization scheme incorporating pipeline desilting exhibits a significant effect on mitigating urban waterlogging, but its effectiveness diminishes with an increase in the recurrence interval. This model demonstrates high accuracy and broad applicability, thereby offering valuable insights for urban waterlogging management.</p>