ResNet-Based Single-Station Localization Method Based on Six-Component Seismometer

Abstract

The adoption of intelligent mining systems improves operation safety and efficiency, requiring accurate localization of shearers in underground environments. However, traditional GPS-based localization methods are ineffective due to underground conditions. Alternative technologies often require receiver arrays or additional signal generators, which can suffer from unreliable connections and extensive hardware deployment. To address these challenges, we propose a novel single-station method that utilizes seismic signals generated by the shearer. This method simplifies the requirements for signal source by analyzing continuous oscillations produced during shearer operation and the hardware implementation using a single six-component (6-C) seismometer to record these oscillations as time-series data. Given the presence of ambient noise, we employ a residual network (ResNet) to extract relevant features from the seismic data. As a deep learning architecture, ResNet incorporates residual blocks to mitigate the vanishing gradient problem, enhancing feature extraction performance. The experimental results demonstrate the effectiveness of the proposed method. Using oscillation data collected during shearer operation to train the ResNet model, we achieve a prediction accuracy of 94.89%, outperforming traditional single-station methods such as cross correlation (CC) and 6-C multiple signal classification (MUSIC). This study demonstrates that our method does not require signal generators or extensive sensor deployments. By leveraging seismic data and deep learning techniques, significant enhancements in shearer localization can be achieved, contributing to the development of safer and more efficient mining operations.