基于机器学习的通风网络故障诊断方法研究  

Research on fault diagnosis method of ventilation network based on machine learning

作　　者：张浪[1,2,3] 张迎辉 张逸斌 李左 ZHANG Lang;ZHANG Yinghui;ZHANG Yibin;LI Zuo(China Coal Research Institute,Beijing 100013,China;Mine Safety Technology Branch,China Coal Research Institute,Beijing 100013,China;State Key Laboratory of Coal Mining and Clean Utilization,Beijing 100013,China)

机构地区：[1]煤炭科学研究总院,北京100013 [2]煤炭科学技术研究院有限公司安全分院,北京100013 [3]煤炭资源高效开采与洁净利用国家重点实验室,北京100013

出　　处：《工矿自动化》2022年第3期91-98,共8页Journal Of Mine Automation

基　　金：煤炭科学技术研究院技术创新基金项目(2020CX-Ⅱ-21)。

摘　　要：机器学习算法通过对已知数据的学习来预测未知数据,现有通风系统故障诊断方法大多针对1种机器学习算法进行研究,无法保证所选算法为最优。针对该问题,对8种机器学习算法进行比较,并选择支持向量机(SVM)、随机森林和神经网络3种算法进行通风网络故障诊断研究。根据矿井通风系统实际布局,按照几何相似、运动相似、动力相似准则构建通风网络管道模型,得到由管道网络分支和管道网络节点组成的通风网络,通过实验获取风量数据,并采用标准化方法对数据进行预处理;通过交叉验证和网格搜索对基于SVM、随机森林、神经网络的通风网络故障诊断模型进行参数寻优。实验及现场测试结果表明,基于SVM、随机森林、神经网络的通风网络故障诊断模型在实验平台测试集上的准确率分别为0.89,0.88和0.95,在煤矿现场测试集上的准确率分别为0.86,0.90和0.96,神经网络模型的故障诊断效果均为最佳。将煤矿现场收集的120组新风量数据输入神经网络模型进行预测,故障诊断准确率达0.98,验证了基于神经网络的通风网络故障诊断模型的可行性和准确性。The machine learning algorithm predicts unknown data by learning known data. Most of the existing fault diagnosis methods of ventilation system focus on a machine learning algorithm, which can not guarantee the selected algorithm to be optimal. In order to solve this problem, eight machine learning algorithms are compared, and three algorithms, support vector machine( SVM), random forest and neural network, are selected to study the fault diagnosis of ventilation network. According to the actual layout of the mine ventilation system, a ventilation network pipeline model is constructed according to the criteria of geometric similarity,motion similarity and dynamic similarity. A ventilation network consisting of pipeline network branches and pipeline network nodes is obtained, and air volume data is obtained through experiments, and the data is preprocessed by a standardized method. Through cross-validation and grid search, the parameters of ventilation network fault diagnosis model based on SVM, random forest and neural network are optimized. The results of experiment and field test show that the accuracy of ventilation network fault diagnosis model based on SVM,random forest and neural network are 0.89, 0.88 and 0.95 respectively on the test set of experimental platform,and 0.86, 0.90 and 0.96 respectively on the test set of coal mine field. The neural network model has the best fault diagnosis effect. 120 sets of fresh air volume data collected in coal mine field are input into neural network model for prediction, and the fault diagnosis accuracy rate reaches 0.98, which verifies the feasibility and accuracy of the ventilation network fault diagnosis model based on neural network.

关 键 词：矿井通风 故障诊断 机器学习 支持向量机 随机森林 神经网络 交叉验证 网格搜索 

分 类 号：TD724[矿业工程—矿井通风与安全]