基于动态特性描述的变量加权型分散式故障检测方法

Dynamic Feature Characterization Based Variable-weighted Decentralized Method for Fault Detection

现代工业生产过程往往具有复杂的动态特性:不同测量变量间会存在不同的时序相关性,且变量间的相互影响会反映在不同的采样时刻上.现有的动态过程监测模型往往不能充分挖掘变量间的动态特性,其故障检测效果也有待进一步提高.在此背景下,本文提出一种基于动态特性描述的变量加权型分散式故障检测方法.利用最大相关最小冗余(Minimal redundancy maximal relevance,mRMR)算法更准确地描述动态过程变量间的相关性关系,并利用该相关性的值对原始增广矩阵进行加权处理,且不同延迟变量对当前测量值的影响大小就通过权值来体现,因此能更加全面地刻画该测量值的动态特性.最后建立一种融合mRMR算法,贝叶斯推理以及动态主成分分析(Dynamic principal componemt amalysis,DPCA)模型的新的分布式建模策略,提高了模型的容错能力和泛化能力,取得了更好的故障检测结果.

Modern industrial process is often accompanied by complicated dynamic behaviours:different measured variables have different serial correlations,and the interactions among these variables are reflected in different sampling instants.The existing dynamic process monitoring models can not fully excavate the dynamic characteristics among variables,then the corresponding fault detection results need to be further improved.Against this background,this paper proposes the dynamic feature characterization based variable-weighted decentralized method for fault detection,which uses the minimal redundancy maximal relevance(mRMR)algorithm to get more accurate description of the relationship among different measured variables in dynamic process,then the augmented matrix can be weighted according to the mRMR values,and the influences of different delay variables on current measured one are reflected by the weights.Hence,the dynamic characteristics of the measured variable can be described more comprehensively.Finally,a new distributed modeling strategy combining the mRMR algorithm,Bayesian inference and dynamic principal component analysis(DPCA),which improves fault tolerance and generalization ability of the model,thus obtaining better fault detection results.