Prediction of Grade Classification of Rock Burst Based on PCA-SSA-PNN Architecture

Abstract

The uncertainty and complexity of rock burst brings great difficulties to the prediction of rock burst grades. In order to estimate the risk grades of rock burst, an integrated method combining principal component analysis (PCA) and sparrow search algorithm (SSA) with probabilistic neural network (PNN) was proposed. Considering that the in situ stress of rock mass, the strength of rock, and the strength of rock mass are the key influencing factors of rock bursts, the maximum in situ stress ${\sigma }_{\max }$ , maximum tangential stress ${\sigma }_{\theta }$ , rock strength ${\sigma }_{\text{ci}}$ , rock mass strength ${\sigma }_{\text{cm}}$ , and three rock burst evaluation indexes ( ${\sigma }_{\theta }/{\sigma }_{\text{ci}}$ , ${\sigma }_{\text{ci}}/{\sigma }_{\max }$ , and ${\sigma }_{\text{cm}}/{\sigma }_{\max }$ ) were selected to constitute the rock burst grade evaluation index system. Forty-three groups of rock burst engineering data were gathered. After preprocessing the rock burst data using PCA, four of the new linearly independent indexes PCA1, PCA2, PCA3, and PCA4 were obtained for estimating rock burst grades. The SSA was utilized to optimize the smoothing factor in the PNN. Using PCA-SSA-PNN-based architecture, a new multi-index rock burst grade prediction method was proposed. The results from the new multi-index rock burst grade prediction method were compared with those from single- and multi-index prediction methods. It shows that the predictions from the multi-index rock burst prediction methods are closer to the actual rock burst grades than that from the single-index rock burst prediction methods; compared with other multi-index rock burst prediction methods, the prediction accuracy of PCA-SSA-PNN is greater (up to 90%) and more available in the prediction of rock burst grades. The results presented herein may provide reference for the rock burst warning.