<p>As traffic simulation platforms play an increasingly important role in traffic management and infrastructure design, the calibration of parameters within microscopic traffic models embedded in these platforms becomes crucial. The inherent physical significance of model parameters and the uncertainty introduced by noise in the data make this calibration process particularly complex. This paper addresses the calibration problem of car-following models, framing it as an optimization problem. Given the non-convexity and non-linearity of the problem, coupled with the high computational cost of large-scale simulations, we propose an improved stochastic fractal search (SFS) algorithm. This algorithm incorporates four enhancement strategies: chaotic mapping initialization, dynamic adjustment of population size, and the Lévy flight diffusion process, among others, balancing exploration and exploitation ability, to improve convergence and robustness. Then, the proposed algorithm is applied to calibrate three typical car-following models across three datasets. We compare its performance with variants of the SFS algorithm, common calibration-solving algorithms, and advanced numerical optimization methods. The results demonstrate that our algorithm is more effective in terms of convergence and resource utilization, and it has a higher probability of finding the global optimal solution. To verify engineering applicability, we further apply ISFS-calibrated IDM parameters to multi-vehicle platoon simulations under dynamic disturbances. Tests show the calibrated parameters ensure stable platoon operation without oscillation, with rapid recovery after disturbances, confirming excellent dynamic stability and anti-interference ability.</p>