Particle swarm optimization (PSO) is a well-known iterative algorithm commonly adopted in wavefront shaping for focusing light through or inside scattering media. The performance is, however, limited by premature convergence in an unstable environment. Therefore, we aim to solve this problem and enhance the focusing performance by adding a dynamic mutation operation into the plain PSO. With dynamic mutation, the "particles," or the optimized masks, are mutated with quantifiable discrepancy between the current and theoretical optimal solution, i.e., the "error rate." Gauged by that, the diversity of the "particles" is effectively expanded, and the adaptability of the algorithm to noise and instability is significantly promoted, yielding optimization approaching the theoretical optimum. The simulation and experimental results show that PSO with dynamic mutation demonstrates considerably better performance than PSO without mutation or with a constant mutation, especially under a noisy environment.