Performing multi-objective optimization for actual public building design has become one of the most challenging subjects in buildings energy efficiency area. Gymnasium is a large energy consumer in public buildings. This study efforts to put forward a novel approach to tackle multi-objective optimization problems for building performance of Qingdao University (QUT) Gymnasium using a new metamodel method. For this purpose, the Nondominated Sorting Genetic Algorithm-II (NSGA-II) was dynamically combined with Multilayer Perception Artificial Neural Network (MLPANN) metamodel, which was previously trained with the co-simulation results conducted using EnergyPlus and Eppy. The new research method also proposes an optimal algorithm coupling Latin Hypercube Sample (LHS) with Principal Component Analysis (PCA) to minimize the total training samples, and guarantees the accuracy of optimization results. The most influential design factors like internal and external wall types, roof types, solar absorptance, windows shading as well as night ventilation (NV) strategy and displacement ventilation (DV) air conditioning system of the gymnasium were considered in three cases of 4×10⁸ possibilities to obtain the optimal trade-off results (Pareto front) between energy consumption and thermal comfort. Finally, a normalized minimum distance decision method was adopted to choose the optimal design configuration from the Pareto front. The optimization results of the study cases showed that reductions were achieved not only in the normalized objectives (88.0% less f_h and 85.3% less f_c) but also in the sub-objectives: up to 78.2% fewer heating energy and 71.3% fewer cooling energy in air conditioning seasons, and up to 97.7% less heating degree-hours and 99.2% less cooling degree-hours in naturally-ventilated seasons, compared to the original configuration by using optimal design takes simultaneous advantage of NV and DV strategies. The method was confirmed to be an efficient and robust tool for gymnasium design, it could reduce the calculation time of whole optimization process from 10 months to 2 days.

对实际公共建筑设计进行多目标优化已成为建筑节能领域最具挑战性的课题之一。体育馆是公共建筑中的耗能大户。本研究致力于提出一种使用新元模型方法解决青岛大学（QUT）体育馆建筑性能的多目标优化问题的新方法。为此，非支配排序遗传算法-II (NSGA-II) 与多层感知人工神经网络 (MLPANN) 元模型动态结合，该元模型之前使用 EnergyPlus 和 Eppy 进行的联合仿真结果进行了训练。新的研究方法还提出了一种优化算法，将拉丁超立方样本（LHS）与主成分分析（PCA）相结合，以最小化总训练样本，并保证优化结果的准确性。在4×10的三个案例中考虑了体育馆内外墙类型、屋顶类型、太阳能吸收率、窗户遮阳以及夜间通风（NV）策略和置换通风（DV）空调系统等影响最大的设计因素。在能耗和热舒适性之间获得最佳权衡结果（帕累托前沿）的^{8 种}可能性。最后，采用归一化最小距离决策方法从帕累托前沿中选择最优设计配置。研究案例的优化结果表明，不仅在归一化目标（f _h减少 88.0% ，f _c减少 85.3%）方面实现了减少)，但在子目标中：空调季节制热能耗降低78.2%，制冷能耗降低71.3%，自然通风季节制热度小时降低97.7%，制冷度小时降低99.2% ，与原始配置相比，通过使用优化设计同时利用了 NV 和 DV 策略。该方法被证实是一种高效、稳健的体育馆设计工具，可以将整个优化过程的计算时间从 10 个月减少到 2 天。