基于PCA-RFR的传感器故障定位方法

Fault isolation method based on random forest regression

摘要:

基于数据驱动的故障诊断方法近些年来得到广泛的研究和应用,但这些方法主要针对于故障检测,对于故障根源的定位尚未得到充分解决。本文提出一种基于主成分分析(PCA)和随机森林回归(PFR)的因果分析故障定位方法(PCA-PFR)。该方法通过将离线故障数据段中的变量作为输入,与之对应的统计量作为输出建立随机森林回归模型,然后通过模型的变量重要性度量来得到过程变量对统计量的因果关系系数,其中值越大的变量被认为越有可能是引起故障发生的故障变量。最后通过一个数值案例和TE过程仿真实验,表明该方法的有效性。

Abstract:

Data-driven fault dignosis methods have been widely studied and applied in recent years, but these methods are mainly focus on the fault detection, and how to determine the fault source has not been fully solved. This paper presents a causal analysis fault location method(PCA-PRF) based on principal component analysis(PCA) and random forest regression(RFR). In this method, the variables in the off-line fault data segment and the corresponding statistics are treated as the input and the output, respectively, and then the causal coefficient from process variables to statistics can get according to the importance measurement of variables based on the random forest regression model. The variable with larger importance measurement is considered to be more likely fault. A numerical case and the Tenessee Eastman Process(TEP) simulation experiment are employed to demonstrate the application of the proposed method, shows the effectiveness of this proposed method.)