Predicting Short-Term Rockburst Using RF–CRITIC and Improved Cloud Model.

Rockburst is a common ground pressure disaster in underground geotechnical engineering. The frequent occurrence of rockburst hazards severely threatens the security of construction workers and facilities on site. Considering the shortcomings of the existing research methods, such as low prediction accuracy, poor interpretability and ignoring the weights of input parameters, this study proposes a new method for predicting short-term rockburst using RF–CRITIC and an improved cloud model (RCICM). Of these, random forest (RF) and CRITIC algorithms were used for optimizing the weights of the prediction indicators, and the cloud model was improved in three aspects: numerical characteristics correction, membership calculation optimization and validity inspection under membership boosting. Six microseismic (MS) parameters were used as inputs to the model, including the accumulative quantity of MS events, accumulative MS energy, accumulative MS apparent volume, event rate, energy rate and apparent volume rate. A random sampling method was used to divide the 105 MS rockburst samples into a training set and a test set, and the model was trained and evaluated separately. The model achieved prediction accuracy of 85.7% on the independent test set, an improvement of 14.3% over the original cloud model. The model was then validated on five rockburst cases at the Asher Copper Mine, the New Jersey Hydroelectric Project tunnels and the Qinling Water Conveyance Tunnel, and correct predictions were obtained. The model has optimal predictive capability compared to existing machine learning models. The sensitivity analysis results revealed the input parameters' influence on the rockburst level and the applicability of the RCICM model with different input parameters. The proposed model has excellent accuracy and applicability and can offer valuable guidance for the safe construction of underground geotechnical engineering. [ABSTRACT FROM AUTHOR]