Intelligent recognition of electromagnetic radiation precursory signals of rock fracture based on random forest-adaptive boosting

Electromagnetic radiation, as a non-contact and real-time monitoring technology, has been widely used in coal rock fracture and rock burst disaster monitoring during coal mining. Electromagnetic radiation signals can reflect the loading state and fracture degree of coal rocks. However, when electromagnetic radiation is used to predict rock fracture, the current study mainly focuses on the trend change of signal strength, counts and other statistical indicators. There is a lack of research on rock fracture prediction based on the electromagnetic radiation signal itself. Therefore, experiments on monitoring electromagnetic radiation in uniaxial compression of rocks were carried out. Differences in the features of ordinary signals (corresponding to the 0–80% σ stage) and precursory signals (corresponding to the 80% σ—σ stage) of electromagnetic radiation during the loading process of rocks were analyzed. The results showed that different signal features distinguished the two types of electromagnetic radiation signals to different degrees. Automatic ranking of feature importance can be achieved by the random forest method. Adaptive boosting method was used to establish intelligent recognition models for two types of electromagnetic radiation signals. The model recognition accuracy was also analyzed when the feature sets were different. It was finally determined that the highest recognition accuracy (92.25%) of the intelligent recognition model for the two types of signals was achieved when the combination of four features was used as the feature set. The research results provide new ideas and methods for the rock fracture prediction using electromagnetic radiation.