Data-physics fusion-driven intelligent early warning of spontaneous coal combustion:From laboratory calibration to domain adaptive field model

LUO Zhenmin; ZHANG Lidong; ZHANG Hui; SU Bing; WANG Tao; WANG Yachao

doi:10.13247/j.cnki.jcumt.20250499

Abstract：Accurate early warning of spontaneous coal combustion is the priority and difficulty of coal mine fire prevention and control.However,the traditional empirical model based on fixed thresholds has insufficient generalization capability due to the difficulty of adapting to the differences in coal quality and environment in different mining areas.To address this challenge,an intelligent early warning model for spontaneous coal combustion across mining areas that integrates data and physical drivers was proposed in this work.At first,the stage characteristics of CO,C2H4,and C2H2 as marker gases were clarified based on 102 sets of coal spontaneous combustion temperature programming experiments.And coal quality and environmental weighting factors were introduced to construct a four-level warning system for spontaneous coal combustion across mining areas with dynamically adjusted warning thresholds.On this basis,a data-physics fusion-driven early warning model was established by transforming the physical laws of coal oxidation process into learning fea-tures.The results show that among various algorithms,the random forest model exhibited optimal prediction performance(training:R2=0.99;testing:R2=0.93)and computational efficiency(prediction time<0.4 s)on experimental data.When directly applied to the Dafosi coal mine(fire scenario),the model accurately output high-temperature warning signals predominantly in orange and yellow.In contrast,at the Huibao coal mine(a safe scenario),it consistently maintained green safety warnings without any false alarms.These results validate that the early warning system and model developed in this study possess excellent cross-mining area adaptability and generalization ca-pability,providing an effective solution to the challenge of developing universally applicable predic-tion models for coal spontaneous combustion.