Abstract: To address the challenges of strong power-frequency interference, acoustic aliasing, and high localization latency in acoustic detection of inter-turn short circuits in dry-type air-core reactors, this paper proposes an intelligent localization method integrating harmonic acoustic fingerprint features with an improved DAMAS2 deconvolution beamforming algorithm. First, a multi-physics coupling model encompassing electromagnetic, mechanical, and acoustic fields reveals the 100 Hz±10 Hz characteristic acoustic emission mechanism induced by eddy currents in short-circuit rings. Second, a 56-channel spiral microphone array (1.5 m diameter) is implemented alongside a focused grid screening deconvolution beamforming (FGS-DAMAS) algorithm, achieving a 72.6% improvement in computational efficiency. Experimental results demonstrate that under 105 dB(A) power-frequency noise, the proposed method attains a localization accuracy of ±0.11 m(95% confidence interval), reduces false alarm rates by 65.8% compared to traditional DAMAS2, and shortens response time to 2.3 s. Validated in ±800 kV converter stations, the method achieves a minimum detectable short-circuit turn ratio of 0.28%.