D-vine copulas混合模型及其在故障检测中的应用

发布时间：2018-02-28 11:44

本文关键词： 过程监控非线性非高斯相关性分析 D-vine copulas　出处：《化工学报》2017年07期 　论文类型：期刊论文

【摘要】：过程监控技术是保证现代流程工业安全平稳运行及产品质量的有效手段。传统的过程监控方法大多采用维度约简方法提取数据特征,且要求过程数据必须服从高斯分布、线性等限制条件,对复杂工况条件下发生的故障难以取得较好的检测效果。因此,提出了混合D-vine copulas故障诊断模型,在不降维的情况下直接刻画数据中存在的复杂相关关系,构建过程变量的统计模型实现对存在非线性与非高斯性过程的精确描述。通过EM算法和伪极大似然估计优化混合模型参数,然后结合高密度区域(HDR)与密度分位数法等理论,构建广义贝叶斯概率(GBIP)指标实现对过程的实时监测。数值例子及在TE过程上的仿真结果说明了该混合模型的有效性及在故障检测中的良好性能。
[Abstract]:Process monitoring technology is an effective means to ensure the safe and stable operation and product quality of modern process industry. Most of the traditional process monitoring methods use dimensionality reduction method to extract data features, and the process data must be distributed from Gao Si. It is difficult to obtain better detection results for faults under complex working conditions under linear and equal limiting conditions. Therefore, a hybrid D-vine copulas fault diagnosis model is proposed, which can directly depict the complex correlation in the data without reducing the dimension. The statistical model of process variables is constructed to accurately describe the nonlinear and non-#china_person0# processes. The mixed model parameters are optimized by EM algorithm and pseudo maximum likelihood estimation, and then combined with the theory of high density region (HDR) and density quantile method. The generalized Bayesian probabilistic GBIPs are constructed to realize the real-time monitoring of the process. Numerical examples and simulation results on te process show the effectiveness of the hybrid model and its good performance in fault detection.
【作者单位】：华东理工大学化工过程先进控制和优化技术教育部重点实验室;
【基金】：国家自然科学基金项目(21406064,21676086) 上海市自然科学基金项目(14ZR1410500)~~
【分类号】：TP277;TQ02

【相似文献】