Advancing Safety in Mining: Machine Learning Approaches for Predicting and Classifying Seismic Bump-Associated Hazardous States

Abstract

The foundation and presumption of underlying risk management in underground coal mines is hazard identification. Even though hazard identification techniques used in underground coal mines have been extensively studied, there is still room for improvement. Because they are experience-based or limited to a single incident or event, traditional hazard identification techniques lack a systematic and all-encompassing identification framework. The material offered explores the intricate problem of predicting high-energy seismic bumps in coal mines that are more than 10^4 Joules. The study uses 2 single predictive models (Random Forest (RF) and Support Vector Classification (SVC)) along with 2 optimization strategies (Artificial Hummingbird Algorithm (AHA) and Turbulent Flow of Water-based Optimization Algorithm (TFWOA)) to tackle this problem. These techniques are applied to improve forecast accuracy. Once the dataset has been divided into hazardous groups and those that are not, a careful analysis of the numerical results is carried out. After a thorough analysis, the most efficient model is the RFC + TFWOA (RFTF) model, which uses Random Forest Classification (RFC) optimized by Turbulent Flow of Water-based Optimization. Notably, the RFTF model attains an astounding accuracy of 0.898 throughout the training phase. This result demonstrates that the RFTF model is more effective than other models at correctly classifying states as hazardous or non-hazardous.

Keywords

• risk management
• seismic hazard prediction
• underground coal mines
• hazardous and non-hazardous states classification