目前燃烧仿真软件中的湍流燃烧模型能够对多种典型燃烧室开展仿真计算，但其预测精度受模型参数影响较大，工程实践中往往需要进行多次的模型参数的调整才能实现仿真结果的优化。如何在湍流燃烧多模型耦合计算中准确量化各模型参数对仿真结果的影响，进而实现对模型参数的校准优化，成为燃烧仿真软件研制面临的瓶颈问题。基于此，该研究选取具有预混燃烧典型特征的非受限强旋流贫油预混火焰，基于活性子空间方法对现有湍流燃烧模型的仿真误差进行了分析，结果表明火焰加厚因子最大值是该旋流预混火焰温度和轴向速度误差的主控模型参数。进而根据活性子空间方法的降维结果，构建了多维参数输入空间的低维响应面，并提出一套基于活性子空间方法的多模型参数优化流程，基于该优化流程对湍流燃烧模型关键参数进行校准，使用简单遗传算法单独对温度或速度进行优化时，关键截面温度平均误差和轴向速度平均误差分别降低7.58%、42.60%；采用带精英策略的非支配排序遗传算法同时对温度和速度进行优化时，关键截面温度平均误差和轴向速度平均误差同时降低1.08%和2.96%，这实现了预混火焰温度和轴向速度仿真精度的大幅提升。

Objective: Turbulent combustion models embedded in current combustion simulation software have successfully predicted performance across various combustor configurations. However, their accuracy is highly sensitive to model parameters, which must be iteratively adjusted and optimized in engineering practice to accommodate diverse combustor geometries, fuel types, and operating conditions. Thus, analyzing the main control mechanisms of turbulent combustion and establishing effective model parameter calibration and optimization processes are crucial for enhancing the prediction accuracy and reliability of combustion simulation software. Methods: This study selects an unconfined, strongly swirling lean premixed flame as the research object. It uses the active subspace method to analyze the effects of the key model parameters of the delayed detached eddy simulation turbulence model and dynamically thickened flame combustion model on the simulation errors of flame temperature and axial velocity. By identifying the high-dimensional mapping direction with the maximum gradient and retaining the main influencing directions in the parameter space, this method reduces dimensionality, based on which a low-dimensional response surface can be constructed for the multidimensional parameter input space. Furthermore, this study proposes a multimodel parameter optimization method that combines the active subspace method, a simple genetic algorithm, and the nondominated sorting genetic algorithm Ⅱ (NSGA-Ⅱ). Specifically, in the analysis of the main control mechanisms, the active subspace method is applied to the TECFLAM flame with the typical swirling premixed characteristics of combustors. Simulation calculations are performed to obtain parameters such as the turbulent dissipation coefficient and maximum flame thickening factor under different values. This helps in identifying the main control mechanisms by which these parameters affect simulation accuracy for the target variables like flame temperature and axial velocity, thereby revealing the parameter optimization direction for improving the accuracy of turbulent combustion simulations. Moreover, the proposed multimodel parameter optimization method for combustor simulations is used to optimize seven key turbulent combustion model parameters of the delayed detached eddy simulation turbulence model and dynamically thickened flame combustion model, including the turbulent dissipation coefficient and maximum flame thickening factor for a typical swirling premixed flame simulation case. Results: Results show that (1) the maximum flame thickening factor is the primary model parameter controlling the temperature and axial velocity errors of swirling premixed flame. (2) Calibrating the key parameters of the turbulent combustion model via the optimization process reduces the average temperature error at critical sections by 7.58% and the average axial velocity error by 42.60% when using the simple genetic algorithm alone. When using elitist NSGA-Ⅱ, the average temperature error at critical sections decreases further by 1.08%, and the average axial velocity error reduces by 2.96%. (3) The optimized model parameters significantly enhance simulation accuracy for typical swirling premixed flames, verifying the effectiveness of the proposed method. Conclusions: The proposed multimodel parameter optimization method effectively improves simulation accuracy for typical swirling premixed flames. It is applicable not only to resolve the parameter optimization problems of turbulent combustion models with the swirling premixed flame characteristics but also offers a new approach for circumventing multiparameter optimization issues in more complex two-phase simulations of combustors, including atomization, evaporation, turbulence, and combustion models.