Author(s)

Zhixuan Tian, Baorui Dong

Corresponding Author

Zhixuan Tian

Abstract

Rock burst has become one of the most threatening disasters to the safety production of coal mines which may cause serious casualties and property losses. This paper is based on the background of this problem by constructing multiple signal features combined with deep learning models, to identify and predict interference signals and precursor signals in monitoring acoustic emission (AE) and electromagnetic radiation (EMR) signals. We constructed five machine learning models including XGboost, decision tree, random forest, Adaboost, and GBDT, classification and recognition can be carried out. Due to the typical differences in data characteristics between precursor signals and interference signals, this article will reuse the model to select feature indicators with high importance rankings. We constructed feature indicators from multiple perspectives including time domain, frequency domain, and seasonality, and screen them. Then, construct multiple machine learning classification models to solve the problem, aiming to obtain the optimal prediction model to achieve maximum recognition accuracy.

Keywords

Feature extraction, Data preprocessing, XGBoost, SHAP, Signal extraction