Abstract:
To address the lack of an intelligentization dimension, a high degree of subjectivity, and poor reliability in traditional exogenous mine fire risk assessment, a model integrating ensemble learning, the Analytic Network Process Incorporating Confidence-Based Triangular Fuzzy Numbers and Entropy Smoothing (CTF-ES-ANP), and combined weighting was proposed. First, an initial indicator system for exogenous mine fire risk assessment was established within the human–machine–environment–management framework, incorporating intelligentization factors. Based on expert questionnaires, a data-driven method combining ensemble learning and statistical inference was used to make the initial indicator screening process more objective and identify key indicators. Second, a data-driven strategy was used to construct the ANP network structure, and CTF-ES-ANP was used to calculate the subjective weights of key indicators to mitigate extreme weight distributions, reduce the risk of low weights approaching zero, and address insufficient quantification of the fuzziness of expert judgments. Next, objective weights were calculated using composite indicator importance scores, and a game-theoretic combined weighting method was used to optimally integrate subjective and objective weights, reducing the dependence of risk assessment results on experts' subjective experience. Finally, a cloud model was used for exogenous fire risk assessment and risk classification. Application of the model to a case-study mine classified its overall exogenous fire risk as low, consistent with recent on-site inspections. The proposed indicator system and assessment model incorporate intelligentization-related factors and can provide a reference for exogenous fire risk classification in intelligent coal mines.