Practical Models To Distinguish Between Seismic Events And Blast Signals

Abstract

The seismic events are contaminated with the blast/noise signals in microseismic monitoring of the underground excavations, negatively affecting the interpretation and detection of high-stress zones. This study proposes explicit and comprehensible classifiers by hybridizing the principal component analysis (PCA) with genetic programming (GP) and classification and regression tree (CART) algorithms. Six discriminant parameters representing the spectrum and source characteristics of the signals were used as input variables. PCA reduced the problem's dimensionality to two components, which were then fed into GP and CART algorithms as the new input variables. A systematic hyperparameter tuning procedure was employed to find the optimum values of the controlling parameters of the algorithms. The hybrid PCA-GP and PCA-CART classifiers provided practical mathematical equations and tree structures, respectively, capable of distinguishing between the signal types with high accuracy. However, the PCA-GP model outperformed the PCA-CART model based on the performance indices.