Identification of the Water Inrush Source Based on the Deep Learning Model for Mines in Shaanxi, China

Roof water inrush disasters in coal mines present safety risks, so swiftly and accurately identifying the source of inrush water is essential for disaster prevention and control. This research employed Python and the Tensorflow deep learning framework, based on a Deep Feedforward Neural Network (DFNN) and used Swarm Intelligence Optimization Algorithms (SIOA) to create six advanced models for identifying the source of roof water inrush in the Binchang mining area, while the characteristics of the input of the sudden water source were abstracted at multiple levels through the multiple hidden layers of neural networks to improve higher classification precision. To counteract overfitting during model development, the Dropout method was applied to enhance the model's hidden layers. Training of the models involved 80% of the collected sample data, while testing utilized the remaining 20%. The evaluation of training efficiency focused on cross-entropy loss and accuracy as primary measures. The study incorporated data from three major water-bearing strata across seven typical mines in the Binchang area, constructing models based on a dataset of 183 water samples. Routine ion concentration tests were conducted on these water samples, selecting eight chemical indicators K+ + Na+, Ca2+, Mg2+, Cl-, SO42-, HCO3-, TDS, and pH as input features, including essential parameters for accurate water source identification. In addition, the models were rigorously evaluated with multi-faceted indicators, including training time, confusion matrices, precision, recall, F1 scores, and graphs of validation loss and accuracy. Results revealed that the validation accuracy of all six SIOA-DFNN models surpassed 91%, much better than traditional neural networks. Specifically, the GWO-DFNN model achieved an outstanding 95.92% accuracy. This highlights the exceptional precision of the optimization algorithm in identifying the source of roof water inrush in the Binchang mining area. This research highlights the high accuracy, applicability, and robustness of deep learning models based on Tensorflow, proving their effectiveness in identifying the source of water inrush in coal seam roofs. These models offer robust technical support for safety measures in coal mining operations.