一种复杂机电系统LE-SVDD异常监测方法

发布时间：2018-05-11 22:38

本文选题：复杂机电系统 + 异常监测方法　；参考：《振动.测试与诊断》2017年03期

【摘要】：复杂机电系统生产过程监测数据具有明显的高维非线性和复杂分布特点,针对传统的方法难以满足复杂系统异常辨识的要求,提出一种拉普拉斯特征映射-支持向量数据描述(Laplacian eigenmaps-support vector domain description,简称LE-SVDD)的异常监测方法。由于高维特征空间中距离很近的点投影到低维空间后距离应该很近,因此改进的LE方法使用一个有权无向图来描述一个流行,用嵌入的方式找到高维数据的低维嵌入,从而能够发现高维数据内部的地位流行结构。通过标准的田纳西-伊斯曼过程(Tennessee Eastman process,简称TE过程)测试和训练数据进行仿真实验,给出了在非线性特征提取和不同时段异常辨识的准确结果。平均漏报率和误报率都比较低,分别为6.063,6和5.625,3.125,这表明LE-SVDD方法在状态监测中具有良好的非线性和高维数据处理能力,适用于工程系统的监测诊断。
[Abstract]:The monitoring data of production process of complex electromechanical system have the characteristics of high dimensional nonlinearity and complex distribution, so the traditional methods can not meet the requirements of abnormal identification of complex systems. A method of anomaly detection based on Laplacian eigenmaps-support vector domain description, (LE-SVDDD) based on Laplacian feature mapping and support vector data description is proposed. Because the distance between the very close points in the high-dimensional feature space should be very close to the low-dimensional space, the improved le method uses a weighted undirected graph to describe a popular method, and finds the low-dimensional embedding of the high-dimensional data by embedding. Thus, the status of the high-dimensional data within the popular structure can be found. Based on the standard Tennessee Eastman process (te process) test and training data, the accurate results of nonlinear feature extraction and anomaly identification in different time periods are given. The average misinformation rate and false positive rate are low, which are 6.063and 5.6253.125, respectively. This indicates that the LE-SVDD method has good nonlinear and high-dimensional data processing ability in state monitoring, and is suitable for monitoring and diagnosis of engineering systems.
【作者单位】：西安交通大学机械制造系统工程国家重点实验室;新疆大学机械工程学院;
【基金】：国家自然科学基金资助项目(51375375) 机械制造系统工程国家重点实验室(西安交通大学)开放课题资助项目(sklms2015009)
【分类号】：TQ050.7

【相似文献】