摘要
为解决矿井通风系统故障诊断难、单一特征模型误判率高的问题,提升故障识别准确性与实时性,提出1种基于机器学习的矿井通风故障智能诊断方法。首先,通过事故树分析法构建矿井通风故障诊断指标体系,结合自动蝶阀故障模拟试验采集通风系统风量、风压数据,采用交叉验证与网格搜索组合策略,对支持向量机(SVM)、随机森林(RF)及神经网络(NN)3种模型的参数进行优化,并对比不同输入特征(单一风量、单一风压、风量-风压复合)下的模型诊断性能;最后将优选模型应用于陕煤张家峁煤矿3个运输巷的通风故障诊断。研究结果表明:风量-风压复合特征输入的神经网络模型诊断性能最优,其故障识别准确率显著高于单一特征模型,且在张家峁煤矿现场验证中,该模型诊断结果与实际故障情况高度吻合,有效验证了方法的可靠性与实用性。研究结果可为矿井通风系统故障的智能、精准诊断提供技术参考。
In order to address the difficulties in fault diagnosis of mine ventilation systems and the high misdiagnosis rate of single-feature models,and to improve the accuracy and real-time performance of fault recognition,this study proposes an intelligent fault diagnosis method for mine ventilation based on machine learning.Firstly,a fault diagnosis indicator system for mine ventilation was constructed using fault tree analysis(FTA).Airflow and air-pressure data of the ventilation system were collected through fault-simulation tests using automatic butterfly valves.A combined strategy of cross-validation and grid search was adopted to optimize the parameters of three models,namely support vector machine(SVM),random forest(RF),and neural network(NN).The diagnostic performance of these models under different input features(single airflow,single air pressure,and combined airflow-air-pressure)was then compared.Finally,the optimized model was applied to diagnose ventilation system faults in three haulage roadways of Zhangjiamao Coal Mine,Shaanxi Coal Industry.The results show that the neural network model with combined airflow-air-pressure features as input exhibits optimal diagnostic performance,with significantly higher fault identification accuracy than models using single features.Moreover,in the on-site verification at Zhangjiamao Coal Mine,the model’s diagnostic results closely match the actual fault conditions,effectively verifying the reliability and practicality of the proposed method.These findings provide technical references for the intelligent and accurate fault diagnosis of mine ventilation systems.
作者
张逸斌
ZHANG Yibin(China Coal Research Institute,Beijing 100013,China;CCRI Tong'an(Beijing)Inlelligent Control Technology Co.,Ltd.,Beijing 100013,China;Beijing Engineering and Research Cenler of Mine Safely,Beijing 100013,China;State Key Laboratory of Intelligent Coal Mining and Stratum Control,Beijing 100013,China)
出处
《中国安全生产科学技术》
北大核心
2025年第11期159-168,共10页
Journal of Safety Science and Technology
基金
煤炭科学技术研究院有限公司科技发展基金项目(2024QN-14,2023CX-Ⅱ-16,2023CX-Ⅱ-15,2023CX-Ⅰ-16)
关键词
矿井通风
故障诊断
神经网络
事故树
交叉验证
网格搜索
mine ventilation
fault diagnosis
neural network
fault tree
cross-validation
grid search
