Gamma-mixture Bayesian method for anomalous coalmine pressure analysis

In the coal mining industry, the management of mine pressure is paramount for ensuring safety and operational efficiency. Anomalous mine pressure data can be indicative of, for example, roof fall, ground instability and rockburst, pose significant risks to facilities and humans and can lead to costly downtime. Recognizing and responding to these anomalies is crucial, however, collecting labelled data can be challenging and costly in some domains/mines. Thus, a straightforward question is whether it is feasible to utilize labelled data from relevant source domains/mines to identify mine pressure anomalies in the target domain/mine. To address such a problem, this study presents a gamma-mixture Bayesian ( \(\Gamma \) $\mathrm{\Gamma }$ MB) approach, by integrating the gamma mixture model ( \(\Gamma \) $\mathrm{\Gamma }$ MM) and Bayesian model in a transfer learning framework. The method has two main processes: domain adaptation and anomaly recognition. The former relies on the \(\Gamma \) $\mathrm{\Gamma }$ MM learned marginal and conditional distributions, aiming to adapt unlabelled target domain samples to source domains. The latter provides an explainable probability model to distinguish normal and abnormal target pressure data. The \(\Gamma \) $\mathrm{\Gamma }$ MB method demonstrates superior performance to ten machine learning approaches on synthetic data and achieves an 86.7% accuracy on average on real-world datasets.