基于优化有向无环图支持向量机的多变量过程均值异常识别

Mean abnormality identification in multivariate process based on optimizeddirected acyclic graph support vector machine

针对多变量过程均值异常模式类型数量太大、一般模式识别工具难以适应的问题,提出优化有向无环图支持向量机。该方法识别效率高,并通过启发式方法生成优化的拓扑结构,即先根据定义在核空间的模式类型平均差异测度对类型编号排序,再依序提取对应两分类支持向量机组成有向无环图结构,使越易区分类型间的支持向量机越靠上层布置,由此缓解分类误差累积效应和弥补上层出现类型分类容错能力的不足,保证相对较高的总体分类准确度。仿真实验表明,优化有向无环图支持向量机用于多变量过程均值异常模式的识别相比其他几种多分类支持向量机在识别精度和效率上具有综合优势。基于优化有向无环图支持向量机构建了多变量过程均值异常识别模型,并在实际齿轮生产中进行了应用实验,验证了模型的有效性和实用性。

Due to the large quantities of mean abnormality mode＇s type, the general identification tool was difficult to adapt it in multivariate process. Aiming at this problem, an optimizing Directed Acyclic Graph Support Vector Ma- chine （DACr-SVM） was proposed, which had high recognition rate and could generate the topology structure through heuristic approach. According to the average difference measure of mode type in kernel space, the type numbers were sequenced. The corresponding binary support vector machines were extracted from sequence to con- struct the directed acyelic graph structure, which could, assign the easily classified support vector machines to upper layers of the structure. It alleviated the classification error accumulation effect and made up the fault-tolerant ability insufficient of classes appearing in upper layers. Therefore, relatively higher overall classification accuracy was en- sured. The simulation experiment results showed that the proposed optimizing DAG-SVM had comprehensive ad- vantages in mean abnormality mode for recognition accuracy and efficiency than other types of multi-class support vector machines. An identification model for mean abnormalities in multivariate process based on optimizing DAG- SVM was built, and its validity as well as practicality was verified in actual gear production process.