Under the current global trend of energy saving and emission reduction and the"Dual-Carbon Goals"( realizing carbon peak before 2030 and carbon neutrality before 2060 ) in China,architectural design tends to pay attention to the rational use of environment. Indoor thermal comfort directly affects people's activities and health,and is one of the main factors of building energy consumption. This paper takes Jin'an Shopping Mall in Harbin as an example to establish a parametric model. While the wind environment simulation and solar radiation simulation are conducted by CFD and EnergyPlus platforms to evaluate thermal comfort,machine learning and genetic algorithm are used to optimize the form of tall atrium space roof under passive natural ventilation. According to the results obtained after optimization,the thermal comfort time of the shopping mall can be increased by 1288 hours to 1350 hours during the annual operating period. By comparing the thermal comfort time of the two roof forms with optimization,the concave roof is the better choice,and the optimized parameter range of the two roof forms is given. The optimization method proposed is expected to provide theoretical basis and method guidance for the subsequent retrofitting design of the atrium of shopping malls and commercial complexes.