Rapid prediction and optimization of Trombe wall design for rural houses using machine learning in China's climate zones

Trombe Wall (TW), as a passive solar technology, shows great potential in enhancing the energy efficiency and thermal comfort of rural houses. To quantitatively evaluate its application value across different climate zones in China and to optimize relevant design parameters, this paper proposes a Multi-Objective Hyperparameter Optimization with NSGA-II for Artificial Neural Networks (MOHNS-ANN) approach. The study first constructs a building performance database comprising 1600 samples based on rural houses located in seven representative climate zones in China. It then simulates the impact of TW configurations on unit energy consumption (EUI), construction cost (CC), and thermal comfort (PMV). The NSGA-II algorithm is employed to determine the optimal hyperparameters for the ANN, which is used to build a multi-output performance prediction model. The results reveal that in the cold, high-irradiation zone, TW reduces unit energy consumption by up to 172 %, while increasing thermal comfort time by more than 17 %. In the severe cold, high irradiation zone and the hot and humid zone with cold winters, the comfort time is improved by approximately 20 %. In the cold and summer hot zones, the cost of constructing a TW system is only about 20000RMB, accounting for just 10 % of the total farmhouse construction cost, indicating favorable economic feasibility. In contrast, improvements in comfort are relatively limited in the moderate high irradiation zone and hot and highly humid zone. Compared to conventional simulations, the proposed machine learning model predicts up to 40 times faster, enabling rapid rural building performance evaluation.