To address the challenges of source number estimation and the susceptibility of weak sources to masking during multi-point concurrent fault acoustic source localization in traction power supply systems, this paper proposes a novel MUSIC-based localization method employing a spiral array and adaptive eigenspectrum entropy estimation. The proposed method formulates an entropy model based on the global distribution of the eigenspectrum and introduces an eigenvalue squaring enhancement mechanism, which facilitates the precise determination of concurrent source numbers under complex background noise conditions. Simulations replicating colored noise and mechanical vibration environments typical of traction substations demonstrate that the proposed algorithm significantly outperforms the traditional Minimum Description Length (MDL) criterion in both estimation accuracy and convergence speed, particularly under conditions of low signal-to-noise ratio (SNR) and limited snapshots. Furthermore, in wide-dynamic-range scenarios where the weak source constitutes merely 20% of the total signal power, the method maintains robust and accurate source estimation. Extensive Monte Carlo simulations corroborate that this approach substantially enhances the spatial resolution and localization accuracy for multiple concurrent fault acoustic sources, thereby exhibiting considerable potential for practical engineering applications.