Research on mine electrical resistivity inversion method based on Deep Learning Method

Coal seam floor water hazards, caused by stress changes resulting from coal mining, are a common type of mine water disaster, and their monitoring and prevention are critical for mine safety. The mine resistivity method, a geophysical exploration technique, is widely used for monitoring and detecting such water hazards due to its high sensitivity to water-bearing structures. In practical monitoring, it is necessary to rapidly and accurately invert apparent resistivity data. However, traditional linear inversion methods are prone to local optima, leading to biased results. In contrast, deep learning-based inversion methods utilize data mining to train networks, avoiding reliance on initial models and enabling fast computation of global optimal solutions.

This study constructs a multi-layer convolutional and skip-connected U-Net model to capture resistivity features at different scales. The model is trained and validated using synthetic data to evaluate its inversion accuracy and efficiency in monitoring coal seam floor water hazards. The results show that the U-Net-based inversion method can accurately identify low-resistivity anomalies associated with water hazards in the coal seam floor and quickly achieve the global optimal solution.

The method is further applied to the inversion of resistivity models with complex boundaries to simulate the impact of stress changes caused by coal mining on the formation of floor water hazards. The results demonstrate that this method is several times faster than traditional linear inversion methods, while maintaining high consistency with the actual model. Therefore, this inversion method provides an efficient new tool for monitoring coal seam floor water hazards and holds great promise for advancing technologies in mine water disaster prevention and geological exploration.