基于FCM-ANN的化工储罐异常检测方法研究.pdf

第 34 卷 第 2 期 计算机应用与软件 VoL 34 No . 2 2017年 2 月 Compute Applications and Software Feb . 2017 基 于 FCM -ANN 的化工储罐异常检测方法研究 杨 小 健 朱 月 $ 钱 景 辉 ( 南京工业大学计算机科学与技术学院江苏南京 211816) 摘 要 如何准确地检测出储罐运行的异常状态是工业控制系统的核心问题，针对传统的有监督学习需要大 量学习样本，而无监督学习准确率不足的问题 ,提出一种基于FCM-ANN 的异常检测方法。该方法基于三层结构 模型 ,FCM层不需要任何先验知识，对数据进行初步异常检测，A N N 层对 FCM 层的每个类分别进行神经网络学 习，最后通过A N N 集成得到检测结果。对采集的储罐运行状态数据进行仿真后，结果表明该方法比ANN、FCM 和 Naive Bayes方法有更优的检测性能。 关键词 储 罐 异 常 检 测 FCM A N N 三层结构模型 中图分类号 TP391 文献标识码 A DOI ：10. 3969/j. issn. 1000-386x. 2017. 02.038 RESEARCH ON ABNORMAL DETECTION OF CHEMICA STORAGE TANK BASED ON FCM-ANN Yang Xiaojian Zhu Yue * Qian Jinghui ( School o f Computer Science and Technology, Nanjing Tech University, Nanjing 211816 ，Jiangsu ，China) Abstract How to detect the abnormal state of the operation of the storage tank is the core problem in industrial control systems. Most of the detection methods proposed so far employ a supervised-leaming or unsupervised-learning technique, the former fails to detect unknown anomalies while the latter requires large amounts of learning data. In order to solve the above problems, this paper presents a hybrid algorithm named FCM-ANN which is a mixture of Fuzzy C- Means clustering and Artificial Neural Network. There are three phases involved in the algorithm, in the first phase, namely the FCM layer, FCM algorithm is used to separate the data into several clusters and most of the abnormal data gather together. In second phase, different ANNs is trained based on various clusters and at last neural network ensemble is used to combine the results of different ANNs. Some experiments are conduc