基于SVM-BN的天然气长输管道燃气轮机故障预警方法研究

发布时间：2019-02-12 17:58

【摘要】：燃气轮机是天然气长输管道的核心动力设备,由于其结构复杂、零部件众多、工作负荷大,在运行过程中常发生故障并触发自保护停机,影响了主管线输气的稳定性。为此,针对天然气干线管道燃气轮机及其附属设备的关键部件、关键故障,利用SCADA系统的在线数据,对故障数据进行分析,建立基于支持向量机的在线异常预测模型和基于贝叶斯网络的故障诊断模型;然后通过引入偏离度建立了上述两种模型相结合的故障预警模型,并在天然气干线管道燃气轮机机组上进行了验证。结果表明:(1)所建立的基于支持向量机的预测模型具有较强的泛化能力,能对压力状态做出较为准确的预测,但是随着预测时间的变长,预测精度也会不断降低;(2)通过关注故障诊断模型中各底事件的发生概率,能够为现场维护人员提供优先检查的依据,进而减少设备非计划外停机事故的发生;(3)在燃气轮机润滑油系统实际发生的故障场景进行的应用与验证结果表明,该模型可以准确地对系统故障提前预测并定位,方法可行、有效,有较高的应用价值。
[Abstract]:Gas turbine is the core power equipment of long distance natural gas pipeline. Because of its complex structure, numerous parts and heavy workload, it often breaks down and triggers self-protection shutdown during operation, which affects the stability of main pipeline gas transmission. Therefore, according to the key components and key faults of gas turbine and its auxiliary equipment in natural gas trunk pipeline, the online data of SCADA system are used to analyze the fault data. The on-line anomaly prediction model based on support vector machine and the fault diagnosis model based on Bayesian network are established. Then the fault warning model combined with the above two models is established by introducing the deviation degree and verified on the gas turbine unit of the natural gas trunk pipeline. The results show that: (1) the prediction model based on support vector machine (SVM) has strong generalization ability and can accurately predict the pressure state, but the prediction accuracy will decrease with the increase of prediction time. (2) by paying attention to the probability of each bottom event in the fault diagnosis model, it can provide the basis for the priority inspection for the field maintenance personnel, and then reduce the unplanned outage accident of the equipment; (3) the application and verification results in the fault scene of gas turbine lubricating oil system show that the model can accurately predict and locate the fault of the system in advance. The method is feasible, effective and has high application value.
【作者单位】：中国石油大学机电学院;中国石油管道联合有限公司西部分公司;
【基金】：国家自然科学基金项目“管道泄漏检测中的退化多源信息融合模型研究”(编号:51204203)
【分类号】：TE973

【相似文献】