Field-calibrated ANN-assisted multi-objective optimization of passive thermal retrofit strategies for existing multi-storey residential buildings

Existing multi-storey residential buildings in hot-summer and cold-winter regions often have poor envelope performance, high energy demand, and inadequate indoor thermal comfort. Passive retrofit design must balance energy use, predicted thermal dissatisfaction, and investment. This study develops a field-calibrated ANN-assisted multi-objective optimization framework for a six-storey residential building in Jingzhou, China. A Grasshopper–Honeybee–EnergyPlus model was assessed against winter indoor-temperature measurements from a bedroom and a living room. Sobol analysis informed the reduction of 18 initial retrofit variables to 15 inputs. A database of 1000 Latin hypercube samples was used to train an artificial neural network surrogate, subsequently coupled with NSGA-II and entropy-weighted TOPSIS. The reported compromise candidate reduced energy use intensity (EUI) from 84.95 to 61.97 kWh/(m 2 ·a), a 27.05% reduction, and predicted percentage dissatisfied (PPD) from 66.90% to 53.61%, a relative reduction of 19.86%, at a unit-area retrofit cost of 108.37 CNY/m 2 . Substantial residual predicted discomfort highlights the limits of the passive-only intervention. The framework links measured winter thermal response, sensitivity-informed surrogate modelling, and case-specific retrofit decision support.