Abstract:

Due to the high dimension of fault diagnosis data in chemical process, the fault features are not easy to distinguish, and the SOM network is easy to fall into the problem of local best. A fault diagnosis method based on KFDA and DE algorithm to optimize SOM neural network was proposed. Firstly, Euclidean distance was used to weight the distance between classes, so as to avoid the problem of overlapping of projected data due to the large distance between classes. As a consequence, the fault data samples could obtain better projection effect and optimize the classification performance. Then, the DE algorithm was used to dynamically adjust the weight vectors of SOM neural network, which effectively avoids the problem of falling into local optimum due to the appearance of "dead neurons". The fault data of TE process and PX disproportionation process were tested. The results showed that compared with traditional SOM network, KFDA-DE-SOM algorithm has higher classification diagnosis accuracy and can be effectively applied to the fault diagnosis of the chemical process.

Key words: process control, neural network, optimization design, differential evolution algorithm, fault diagnosis

摘要：

针对化工过程故障诊断数据存在高维度、故障特征不易区分、自组织映射（self-organizing map，SOM）网络易陷入局部最优等问题，提出了一种基于改进核Fisher判别分析（kernel Fisher discriminant analysis，KFDA）与差分进化算法（differential evolution，DE）优化SOM神经网络相结合的故障诊断方法。该方法首先利用欧氏距离对类间距进行加权处理，以避免因类间距离过大造成投影后的数据存在重叠的问题，使故障数据样本获得较好的投影效果，优化分类性能；然后,利用DE算法对SOM神经网络的权值向量进行动态调整，有效避免了由于“死神经元”的出现陷入局部最优的问题；最后，通过对田纳西-伊斯曼（tennessee-eastman，TE）过程和对二甲苯（paraxylene，PX）歧化工艺过程的故障数据进行诊断测试。结果表明，与传统SOM网络相比，提出的KFDA-DE-SOM算法具有较高的分类诊断精度，可有效应用于化工过程的故障诊断。

关键词: 过程控制, 神经网络, 优化设计, 差分进化算法, 故障诊断