为了得到泛化能力高的煤自燃倾向性预测模型，本文基于煤的多指标气体和工业分析参数，采用随机森林、神经网络、支持向量机和Stacking堆叠，通过预测煤的自燃温度和自然发火期，评估煤自燃倾向性。结果表明采用Stacking堆叠方法的预测模型泛化能力最佳，该预测模型特征重要性表明挥发分和乙烯分别在煤自然发火期和煤自燃温度预测中具有最强的关联性。分析模型性能指标，发现增加数据量可以显著提升随机森林、神经网络、支持向量机和Stacking堆叠方法在煤自燃温度预测模型中的泛化性能。对于煤自然发火期预测模型，单纯增加数据量意义有限，需要探索更多新特征以提升模型性能。本文对煤自燃倾向性的预测和模型优化提供了一些有价值的参考。

In order to develop a high-performance model for predicting the spontaneous combustion tendency of coal, this study utilizes multi-indicator gas and industrial analysis parameters of coal. It employs random forest, neural network, support vector machine, and Stacking stacking methods to predict the spontaneous combustion temperature and natural ignition period of coal in order to evaluate its spontaneous combustion tendency. The findings indicate that the Stacking stacking method-based prediction model exhibits superior generalization ability. Furthermore, feature importance analysis reveals that volatile matter and ethylene demonstrate strong correlations with predicting the natural ignition period and spontaneous combustion temperature of coal respectively. Analysis of model performance indicators suggests that increasing data volume significantly enhances the generalization performance of random forest, neural network, support vector machine, and Stacking stacking methods in predicting coal spontaneous combustion temperature. However, simply increasing data volume has limited impact on improving the prediction model for coal's natural ignition period; instead, exploring new features is essential for enhancing model performance. This study offers valuable insights into predicting coal spontaneous combustion tendency and optimizing predictive models.

松散多孔介质自燃过程热质传输与水分迁移耦合机理研究(LY18E040001),浙江省自然科学基金面上项目