Automatic recognition of effective and interference signals based on machine learning: A case study of acoustic emission and electromagnetic radiation

Integrated monitoring technology using acoustic emission (AE) and electromagnetic radiation (EMR) is a promising mean for monitoring coal and rock dynamic disasters. However, due to the complex underground environment, blasting, drilling and mechanical operations may generate interference signals that affect the accuracy of early warning. It is essential to study the automatic recognition of effective and interference signals for AE and EMR. For this reason, a field test of synchronous AE-EMR monitoring was conducted in a coal mine. Further, the time domain, frequency domain and fractal characteristics of effective and interference signals for AE and EMR were analyzed, and sensitive characteristics were obtained by the Relief algorithm. Based on this, automatic recognition models of effective and interference signals were established by Fisher's linear discriminant method, support vector machine and ensemble learning method, respectively. Field application showed that support vector machine performed higher recognition accuracy. The AE and EMR warning indicators were obtained based on the signal recognition model, and the results show that the indicators reflect the risk of rock burst more accurately and earlier than the original indicators before the removal of interference signals.