Abstract: Research on optimized design and accuracy enhancement for geosound monitoring systems was conducted by employing the Erlihe Lead−Zinc Mine as a case study to address the frequent occurrence of concealed ground pressure hazards and inadequate adaptability of traditional monitoring methods during deep lead−zinc mining. A geosound sensor network was constructed based on the spatial geometry of deep excavations and in-situ stress environment, covering key sublevels between 900 m and 1 100 m depths. The spatial response characteristics and localization error sensitivity of the sensor layout were systematically analyzed by introducing multiple seismic source simulations with arrival time error perturbations. Furthermore, an improved A* algorithm combined with GPU parallel computing and a velocity-independent source localization method was proposed to achieve efficient seismic source identification and ensure accuracy in complex wave propagation environments. The results indicate that under the arrival time perturbations with a standard deviation up to 0.2 ms, the localization error of the optimized monitoring system is controlled within 6 m, demonstrating the system's strong robustness against noise and three-dimensional coverage. Compared to the fixed velocity method （average localization error of 84.8 m） and MSLM−WV （average localization error of 46.4 m）, while substantially improving computational efficiency, the ISACE−GPU method significantly reduces the average localization error to 25.8 m and ensures localization accuracy. The proposed optimization strategy and validation system provide a technical basis and transferable method for the engineering deployment and performance evaluation of geosound monitoring systems in deep mines.