为解决矿井通风系统故障分支判识不准确的问题，引入单分类算法，构建了多个单分类支持向量机(One-Class Support Vector Machines，OCSVM)集成的通风系统故障诊断模型。模型采用统一超参数并设计了尺度统一公式以实现多个输出尺度的统一，将通风系统故障诊断问题转变为最大决策距离问题，建立仅需正常样本参与训练的通风系统故障诊断半监督学习模型，实现对矿井监测风速数据的有效利用。进行了 KEEL公开数据集和东山煤矿生产矿井实例试验，结果表明，单分类集成模型能够解决多分类问题，与其他单分类集成模型相比，单分类支持向量机集成(Multi-Class One-Class SVM。MC-OCSVM)模型具有最佳的泛化性，所提模型能够快速准确地识别通风系统故障分支，故障诊断准确率达93。2％，单次故障诊断时间为1。2 s，具有较强的鲁棒性。研究工作是实现矿井通风智能化的基础，为通风系统故障诊断提供技术支撑。
MC-OCSVM model for intelligent fault diagnosis of mine ventilation system
To solve the problem of inaccurate fault branch identification in mine ventilation system,a fault diagnosis model of Multi-Class One-Class Support Vector Machines(MC-OCSVM)for ventilation system is constructed by introducing single classification algorithm.An One-Class SVM(OCSVM)model is trained separately for each branch in the mine ventilation system,and a multi-classification model is obtained by combining the OCSVM set of all branches.After the test sample is input into MC-OCSVM,each OCSVM will output a predicted value.The predicted value represents the distance between the test sample and the normal sample classification decision vector.The branch corresponding to the maximum predicted value is the fault branch.In order to avoid the problem that the output value of OCSVM cannot be compared.The model adopts unified hyperparameters and designs a unified scale meter.It realizes the unity of multiple output scales.This method transforms the ventilation system fault diagnosis into the maximum decision distance.It also realizes the effective use of mine monitoring wind speed data.KEEL open data set and Dongshan Coal Mine test results show that:(1)The single classification integrated model can solve the multi-classification problem;(2)Compared with the Multi-Class One-Class Extreme Learning Machine(MC-OCELM)model and Multi-Class Support Vector Data Description(MC-SVDD)model,the MC-OCSVM model has the best generalization and robustness;(3)The MC-OCSVM model can quickly and accurately identify the fault branch of the ventilation system.The fault diagnosis accuracy rate is 93.2％.The values of Precision,Recall,F1,and Gmean are 93.2％.The single time of fault diagnosis is 1.2 s.The model can meet the need of timely and accurate fault diagnosis of ventilation system in large production coal mines.The research is the basis of realizing intelligent mine ventilation and provides technical support for fault diagnosis of ventilation system.
safety engineeringmine ventilationintelligent algorithmfault diagnosisone-class classification integrationOne-Class Support Vector Machines(OCSVM)
NETL
NSTL
万方数据
