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.