A data-driven fuzzy model for prediction of rockburst

Abstract

ABSTRACT Rockburst is the sudden ejection of highly stressed brittle rocks. This can impose major risks on underground projects. Despite all the studies conducted on this phenomenon since the 1900s, it is still fairly challenging, unknown and difficult to predict. As there is a tendency of studying less-known phenomena with intelligent methods (e.g. ANN, Fuzzy, Neuro-Fuzzy, ANFIS, etc.), this study tries to utilise some rather new Data-Driven fuzzy models for rockburst prediction. For this purpose, 174 rockburst case studies were analysed with help of three conventional rockburst prediction criteria as a benchmark. Later, with respect to input data sensitivity to placement order, LoLiMoT and two Takagi–Sugeno Sigmoid-based membership function (TS-SBMF) methods were used for submodel generation in a fuzzy model. Furthermore, by fine-tuning the input parameters of the first-order TS-SBMF approach which proved to be the most promising method, extensive simulations were done in order to check the true performance of the purposed model and make sure of its robustness. Eventually, the purposed models illustrated high average accuracy for both “Train” and “Test” datasets and simultaneously increased the exact predictions and decreased wrong predictions.