GCN-based prediction method for coal spontaneous combustion temperature

Accurately predicting coal spontaneous combustion temperature is crucial for preventing coal mine fires, reducing loss of life, and safeguarding property. Traditional gas concentration prediction methods typically rely on a limited set of gas concentration parameters, often neglecting the complex interactions among them, which impacts prediction accuracy. In this study, we propose a prediction model based on Graph Convolutional Network (GCN), which integrates gas concentration parameters and their interactions to enhance prediction performance. First, the parameters are represented as nodes in a directed graph, with edges defined by the chemical reactions occurring during the coal self-heating process. Then, multiple GCN layers are employed to capture the intricate relationships between the nodes. The model was trained and tested on datasets from multiple coal mines, and the results demonstrate that the GCN model outperforms existing methods. Specifically, for datasets from different coal samples, the MAE values are 2.49, 3.53, and 2.92, while the R² values for all datasets exceed 0.99. This demonstrates that considering the interrelationships between different gas indicators significantly improves the accuracy of coal spontaneous combustion temperature prediction, validating the effectiveness of the proposed method and contributing to reducing the occurrence of coal mine spontaneous combustion disasters.