Abstract
Aiming at the complexity of coal spontaneous combustion prediction, a hierarchical prediction model based on multi-objective genetic NSGA-II optimized random forest algorithm was established to classify the coal spontaneous combustion temperature class by multi-indicator gases, and the coal spontaneous combustion temperature class by multi-indicator gas concentration. In this paper, CO, CO(2), CH(4), C(2)H(6), C(2)H(2), CO(2)/CO, CH(4)/C(2)H(6) are selected as the index gases of coal spontaneous combustion by using the coal natural ignition experiment, summarizing the rule of change between the concentration of each component gas and the coal temperature, and combining with the coal-oxygen composite mechanism, the coal spontaneous combustion and oxidation process is divided into seven phases, including the latent phase, the critical phase, oxidation phase, fission pyrolysis, accelerated oxidation phase, accelerated growth phase, combustion phase, and determine the temperature level of each stage and the index gas characterization parameters of the corresponding level. In this paper, the prediction results of the constructed NSGA-II-RF model are compared and analyzed with those of the PSO-RF, PSO-SVM, and RF models, respectively, and the results show that, in terms of the average accuracy, the model accuracy of the NSGA-II-RF is improved by 3%, 6%, and 9%, respectively, compared with that of the PSO-RF, PSO-SVM, and RF. In this paper, the model precision is calculated and analyzed by four categorical evaluation indexes, namely, accuracy, precision, recall, and F(1) value of each model, and the results show that, in general, NSGA-II-RF > PSO-RF > PSO-SVM > RF. Improvements in safety and operating time in the mining industry can be achieved through modeling, improvements in modeling that can increase the practical value of the research and contribute to real-world mining operations, such as early detection of burn risks and enhanced mine safety protocols.