To address the random and fading noise in a distributed acoustic sensing system, this study proposes a variational mode decomposition (VMD) approach enhanced with permutation entropy and optimized using the starfish optimization algorithm. This method enables automatic parameter selection, thereby overcoming the limitation of manual parameter tuning in conventional VMD. It is further validated on a phase-sensitive optical time-domain reflectometry system for validation. Experimental results demonstrate that, for disturbance signals at different frequencies, the proposed algorithm consistently achieves SNR improvements of exceeding 14 dB, highlighting its robustness and effectiveness in noise suppression.