Building retrofit multiobjective optimization using neural networks and genetic algorithm three for energy carbon and comfort

In response to the urgent need for energy-efficient retrofits in industrial buildings under global climate goals, this study presents a robust multi-objective optimization framework that integrates building performance simulation, surrogate modeling, evolutionary algorithms, and decision analysis. A representative old factory building in Wenzhou, China, is selected as the case study. DesignBuilder is used to simulate energy consumption, thermal comfort, and carbon emissions. To reduce computational costs, surrogate models based on Backpropagation Neural Networks (BPNN) and Support Vector Regression (SVR) are developed and compared in terms of predictive performance.The results show that BPNN demonstrates superior predictive accuracy compared to SVR, with higher R and lower RMSE values. Then, the Non-dominated Sorting Genetic Algorithm III (NSGA-III) is employed to generate a set of Pareto-optimal solutions, and the entropy-weighted TOPSIS method is applied to identify the most balanced retrofit option. The optimized design results in a 10.06% reduction in thermal discomfort hours (Tdh), a 35.45% reduction in energy density index (EDI), and a 28.86% reduction in life-cycle carbon emissions (LCCO₂), respectively. Overall, the proposed framework proves to be highly applicable to the low-carbon renovation of existing industrial buildings, offering a practical and scalable decision-support approach for achieving a balance among energy efficiency, environmental sustainability, and indoor comfort.