To address the challenges of slow convergence and low training efficiency in Markov Chain Monte Carlo (MCMC) methods for high-dimensional parameter fitting and uncertainty quantification, this paper proposes a stepwise parameter freezing strategy to accelerate convergence. The proposed approach leverages the Gelman-Rubin convergence diagnostic to monitor parameter convergence status and employs uncertainty evaluation from the last n iterations to iteratively freeze converged parameters. By reducing redundant sampling and guiding the MCMC process towards more efficient exploration, the method significantly enhances convergence speed and ensures stable and reliable Bayesian parameter fitting in data-driven multi-parameter tasks, as demonstrated by experimental results.