催化裂化反再系统的故障诊断方法研究

Research on Fault Diagnosis of Reaction-Regenerator System of FCCU

摘要:

考虑到石油化工过程系统复杂,变量繁多,非线性关系极强,故障样本数据难于获取,故利用支持向量机对炼油厂催化裂化装置反应再生子系统的故障状态进行模式识别;且支持向量机参数C、σ对分类精度有很大影响,采用了改进的遗传算法对其进行优化；并采用了决策树算法进行多类分类,根据类间分离测度,从最难分类(类间分离测度最小)的两类样本集开始训练,将其合并为一个类簇后同其他样本集一起,再从中寻找最难分类的两个样本集合并,如此逐步合并最终得到训练模型;实验结果表明,利用改进的遗传算法优化惩罚系数C和核函数参数σ后,缩短了分类时间,提高了分类准确率,基于决策树算法的支持向量机能有效地解决一对一和一对多分类算法中存在的无法辨识区域的问题,能很好地识别故障类型,对催化裂化装置的故障诊断有显著的指导作用。

Abstract:

The system of petrochemical process is complex, its variables are various, and their relationship is strongly nonlinear, so use support vector machine to identify the pattern of the fault state on the regenerative subsystem of the catalytic cracking unit in oil refineries. Taking into account the parameters C、sigma have a great influence on the classification accuracy, so an improved genetic algorithm is used to optimize the parameters; And chose decision tree algorithm to classify, according to the separation measure between different categories, start training from the more difficult and most difficult classification sample sets, that is the separation measure is minimum, then combine them into a class cluster and put together with other sample sets, find two of the most difficult classification sample sets and combine them again, combine clusters like this and get the training model finally. The results show that after using the improved genetic algorithm to optimize penalty factor C and the kernel function parameter sigma, the time of classification is shortened and the accuracy of classification is improved. The algorithm based on decision tree support vector machine effectively solve the problem that area cannot be identified that exists in one against one and one against other multiple classification algorithm, can identify the fault types well and has a significant guidance on the fault diagnosis of FCCU.