An efficient hybrid sampling surrogate model-based method for the indoor environment multi-objective optimization

In multi-parameter optimization of indoor environments, achieving a balance between the number of constructed samples and optimization efficiency remains challenging, and optimization performance is often dependent on large datasets. Therefore, achieving satisfactory optimization results with a limited number of samples remains a major challenge in the field of indoor environmental optimal design. To address this, this study proposed a sample construction method integrated with proper orthogonal decomposition (POD), which is further coupled with the Kriging surrogate model to develop two surrogate modeling approaches: multi-fidelity Kriging surrogate model based on POD (MFK-POD) and single-fidelity Kriging surrogate model based on POD (SFK-POD). These two approaches achieve three key improvements. First, MFK-POD attains predictive accuracy comparable to the classical ordinary Kriging (OK) model while reducing computational time by 12.4%, and SFK-POD achieves a 21.7% reduction in computational time, thereby enabling rapid optimization. Second, the study identifies the “moderation principle” in POD-based sample expansion, where setting the expansion ratio n = 1 (i.e., doubling the initial sample size) achieves an optimal balance between efficiency and accuracy. Finally, in the optimization design of a typical indoor environment, MFK-POD achieves a 54.6% reduction in CO₂ target concentration, a 17.3% reduction in energy consumption when compared with the optimization results of existing methods, and a prediction deviation of less than 2%. In addition, the Spacing value of the Pareto solution set is reduced by 70.2%, and the hypervolume (HV) value is increased by more than 1 time. The SFK-POD model is suitable for time-sensitive scenarios requiring lower precision and rapid iteration, while the MFK-POD model is better suited for engineering optimization tasks that demand a balance between accuracy and efficiency. This framework provides a novel methodological basis for high-efficiency and high-precision multi-parameter optimization in complex thermal-humid environments.