A quantitative pre-warning for coal burst hazardous zones in a deep coal mine based on the spatio-temporal forecast of microseismic events

The quantitative prediction for a coal burst is challenging since the coal burst mechanism is extremely complex with a verity of influencing factors involved. This study proposes a data-driven strategy to dynamically determine the coal burst hazardous zones in a deep coal mine based on quantitative predictions for microseismic events. A deep learning model, MSNet, comprising a convolutional module, a recurrent module, a skip-recurrent module, and an autoregressive module is built to predict the time, location, and energy for imminent microseismic events. More than ten thousand microseismic events from a workface were collected to form the database for the MSNet model training and testing. The results indicated that the MSNet can predict the event location accurately but that it predicts event timing less accurately. The MSNet demonstrated the worst prediction accuracy for event energy. Furthermore, this study analyzed the possible causes of the model’s prediction errors and provided ways for enhancing the model’s performance. Finally, a coal burst intelligent pre-warning platform was developed, which has been successfully used in coal mines at present. This study realized the quantitative forecast for coal burst hazardous areas on a preliminary basis while laying a foundation for coal burst timing risk prediction.