An optimized deep neural network for rockburst damage potential modelling

Managing ground prone to rockburst is challenging, especially in seismically active underground mines. Over the past few decades, numerous studies have been conducted on predicting rockburst damage potential. However, in most cases, only fair model performance was achieved due to the complex nature of rockburst as a seismic event and the non-linearity of data. To overcome these limitations, this paper presents a more reliable model for predicting rockburst damage potential (RDP). An Artificial Neural Network was established, and its parameters were optimized using the Adam optimizer. Rockburst data consisting of 254 case histories were compiled and used to model the RDP scale. The dataset was divided into two parts: a training set, which accounted for 80% of the dataset, and a separate test set, which accounted for the remaining 20%. Cross-validation technique was applied to the training set to avoid overfitting. The input parameters for the model included the capacity of the ground support system, stress conditions, presence of geological structure, excavation span, and peak particle velocity. Several performance indices were used to evaluate the model, and the overall results indicate good performance. In conclusion, this study could help engineers adequately assess rockburst damage in seismically active mines.