Multi-objective optimization of rural residential buildings in cold regions using BPNN and NSGA-III: A case study of Shangqiu

This study proposes a multi-objective optimization framework integrating dynamic simulation, a Backpropagation Neural Network, and the Non-dominated Sorting Genetic Algorithm III to improve energy efficiency, thermal comfort, and affordability of rural houses in cold regions. Using a representative dwelling in Shangqiu, Henan, 21 key variables were identified through sensitivity analysis, and 30,000 samples were generated by Latin Hypercube Sampling. The BPNN achieved high prediction accuracy (R 2 =0.967 for EUI th , 0.988 for PPD). NSGA-III optimization produced 100 Pareto-optimal retrofit schemes, among which three representative solutions—comfort-oriented, balanced, and energy-saving—were selected using TOPSIS. Compared with the baseline, energy consumption decreased by 46.4-67.2%, and thermal comfort improved by up to 67.5%, with retrofit costs between ¥36,000-52,000, within the economic capacity of rural households. The findings demonstrate that enhancing envelope insulation and reducing window U-values effectively achieve energy savings and thermal stability, offering quantitative guidance for low-carbon, cost-effective rural housing retrofits.