Large space buildings play a significant role in modern society because of their environmental advantages and market value. While the zonal model is promising for the efficient and rapid evaluation of the stratified thermal environment, there is a lack of a reasonable and convenient zoning strategy with the advent of modern computing. This paper presents a universal and practical zonal model, in which a simplified momentum equation is applied to consider air momentum preservation, transformation, and dissipation. Hence, the zoning structure is generalized and flexible. Moreover, limiting the dimensionless temperature constraint between adjacent zones establishes the connection between thermal nonuniformity and zoning results automatically. Simultaneously, the dimension and number of zones should be restricted within reasonable ranges to satisfy the characteristics of zonal simulation and reach the criteria of convergence. To further explore and validate the zonal model, a reduced-scale experimental model was constructed to replicate the thermal stratification in a mechanically ventilated large space by considering many crucial realistic factors. A particle image velocimetry (PIV) measurement was then conducted to visualize the airflow pattern and support the partition of zones. The results showed that the zonal simulation with adaptive zoning method can realize a similar accuracy with fewer zones and exhibit a better tolerance for zoning results compared to the conventional empirical zoning method. Furthermore, a case study of an atrium was performed to demonstrate the practicality and efficiency of the method for long-term dynamic simulations of complex thermal environments and building energy use.

大空间建筑因其环境优势和市场价值在现代社会中发挥着重要作用。虽然分区模型有望有效、快速地评估分层热环境，但随着现代计算的出现，缺乏合理、方便的分区策略。本文提出了一种通用实用的纬向模型，其中应用简化的动量方程来考虑空气动量的保持、转换和消散。因此，分区结构是通用的和灵活的。此外，限制相邻区域之间的无量纲温度约束会自动建立热不均匀性和分区结果之间的联系。同时地，区域的维度和数量应限制在合理范围内，以满足区域模拟的特点并达到收敛标准。为了进一步探索和验证区域模型，通过考虑许多关键的现实因素，构建了一个缩小比例的实验模型来复制机械通风大空间中的热分层。然后进行粒子图像测速 (PIV) 测量以可视化气流模式并支持区域的划分。结果表明，与传统的经验分区方法相比，自适应分区方法的分区模拟可以在较少的分区下实现相似的精度，并且对分区结果具有更好的容忍度。此外，