为了从石化装置大量工艺监测数据中提取有效的故障特征信息,及时地发现故障并准确地识别故障原因,提出了一种基于PCA和RBF神经网络的故障监测与诊断方法。首先获取工况样本,建立PCA模型,降维提取统计特征;设定正常工况SPE统计量阈值,建立在线工况SPE统计量,由此进行故障监测。然后对故障样本进行PCA降维,构建多个RBF神经网络模型,用以实施在线故障诊断,识别故障原因。最后把某石化公司气体分馏装置脱异丁烷单元作为实例,采用UniSim Design软件对该单元进行过程动态模拟,获得工况监测样本,建立了故障监测与诊断模型。研究结果表明,所提出的方法不仅能有效地对工况进行状态监测,而且能快速和准确地诊断故障。%To extract effective fault feature from a large number of the process monitoring data in petrochemical plant, a fault monitoring and diagnosis approach based on PCA ( Principal Component Analysis) and RBF ( Radial Basis Function) neural network was developed to find the fault timely and identify the failure cause accurately in this paper.First, the obtained data samples was used to establish PCA model, in which the data feature was extracted through dimension reduction. SPE ( Squared Prediction Error) statistic threshold in normal condition was set, and the SPE statistic in real⁃time condition was established, thereby fault monitoring was conducted.Then, by using fault samples with low dimension by PCA, the multiple RBF neural network models were constructed for diagnosing on⁃line fault and identifying fault causes.Finally, a deisobutanizer unit in the gas fractionation plant in a petrochemical company was taken as a study case. Fault monitoring and diagnosis model was constructed with process monitoring samples obtained from the dynamic simulation by the UniSim Design software.Results show that the proposed method not only can effectively monitor conditions, but also can quickly and accurately diagnose the fault.
展开▼