Surrogate assisted optimization of decentralized wastewater facilities in severe cold regions

To address freezing failures in decentralized rural wastewater treatment systems in severe cold regions and reduce the computational burden of design optimization, this study proposes a surrogate-assisted multi-objective optimization framework driven by high-fidelity multiphysics simulations. A three-dimensional transient phase-change heat transfer model of the coupled foundation-soil-equipment system is established to generate training data under extreme winter conditions. Using 2500 Latin Hypercube Sampling points, a Bayesian Optimization-Random Forest (BO-RF) surrogate model is developed and coupled with NSGA-II to optimize life-cycle cost, thermal risk, treatment efficiency, user experience, carbon emissions, and land use. The MARCOS method is applied to rank 200 Pareto-optimal solutions, and a robust shallow-burial design with passive insulation is identified. A case study in Harbin, China, shows that the optimized scheme attenuates more than 85% of external temperature fluctuation. The BO-RF model achieves high accuracy (R²=0.961–0.985). Compared with conventional heuristic strategies, the proposed framework reduces life-cycle cost by 38%, improves user experience by 34%, and shifts the thermal risk indicator from 1.85 to a safe margin of $$-5.5$$, with only marginal increases in carbon emissions and land use. The results offer practical quantitative support for resilient and cost-effective infrastructure deployment in cold regions.