[Objective] Conventional grey pipe-network-dominated drainage systems exhibit limited adaptability when confronted with beyond-design storm events. Existing optimization studies of green-grey infrastructure are predominantly conducted under historical rainfall conditions and insufficiently account for future climate change scenarios and their associated uncertainties，particularly with respect to the systematic selection and integration of multiple climate models. To address these gaps，this study constructs a multi-objective optimization framework coupling an XGBoost surrogate model with the NSGA-III algorithm，driven by CMIP6 multi-model climate projections. [Methods] Dahongmen Area in the Liangshui River Basin of Beijing was taken as a case study. We first evaluated eight CMIP6 global climate models （GCMs） that have shown relatively strong performance in northern China. Using historical rainfall observations from 1982-2014 as the benchmark，spatial downscaling was conducted via linear interpolation，followed by bias correction using the Delta method. A comprehensive assessment framework was then established by integrating the Taylor Score （TS） and the Interannual Variability Score （IVS）. Based on this framework，three models—EC-Earth3，ACCESS-CM2，and IPSL-CM6A-LR—were identified as the best-performing candidates. These selected models were subsequently combined using a weighted ensemble approach to construct future rainfall sequences under SSP1-2.6，SSP2-4.5，and SSP5-8.5. Second，to improve computational efficiency for optimization，an XGBoost （XGB） surrogate model was developed to characterize the nonlinear response relationships among rainfall characteristics，the deployment scale of green-grey infrastructure，and the resulting total runoff and cumulative overflow. Finally，with the minimization of annualized cost，total runoff，and total overflow as the objective functions，the XGBoost was coupled with the NSGA-III algorithm for multi-objective optimization. This produced Pareto-optimal solution sets under each scenario. Representative designs—including cost-optimal，compromise，and benefit-optimal solutions—were then selected to systematically analyze the configuration structure and evolutionary patterns of green-grey infrastructure across varying investment levels. [Conclusions] （1） EC-Earth3，ACCESS-CM2，and IPSL-CM6A-LR show relatively good performance in precipitation simulation for the Dahongmen area of Beijing. Compared with single models，the weighted multi-model ensemble improves the simulation accuracy of historical precipitation and reduces the uncertainty caused by biases of individual models. （2） Under different future emission scenarios，extreme rainfall intensity shows an overall increasing trend. Under the SSP5-8.5 scenario，the 10-year return-period rainfall exceeds the historical 50-year level，and the 100-year return-period rainfall intensity increases by 43.9%，indicating a higher risk of exceedance for urban drainage systems. （3） Optimization of green-grey infrastructure effectively reduces runoff and overflow risks. Under the low-emission scenario （SSP1-2.6），overflow can be completely controlled. Under the high-emission scenario （SSP5-8.5），a certain overflow risk still exists even at high investment levels，and the investment cost required to achieve the same control target increases with the intensification of emission scenarios. （4） The allocation of green-grey infrastructure presents obvious staged evolutionary characteristics. In the early stage of optimization，centralized grey detention facilities are dominant，which enhances the basic regulation capacity of the drainage system. With increasing investment，the optimization strategy gradually shifts toward distributed green infrastructure. Within green infrastructure measures，the priority changes from permeable pavement to green roofs，reflecting a structural transition from centralized detention to source control. This study reveals the adaptive evolution mechanism of urban green-grey infrastructure configuration under climate change，and provides a scientific basis and technical support for the phased construction and investment decision-making of urban drainage systems under intensified extreme rainfall conditions.