Fault Root Diagnosis of Industrial Process Based on Random Forest-Partial Symbol Transfer Entropy

Fault diagnosis is an important mean to protect the safety of industrial processes. In recent years, the causal analysis method based on data driven is widely used in industrial process of the fault root diagnosis, such as the Transfer entropy(TE), which is able to quantify information transfer between two time series and get a direct causal relationship between two variables. Partial Transfer Entropy (PTE) is a multi-dimensional extension of TE, only recognizes the direct causal relationship. But PTE method is not suitable for the non-stationary process, and the computational complexity is very high, so this paper proposes a Random Forest-Partial Symbols Transfer Entropy method (RF-PSTE) to identify the network of causal relationship. It is not only suitable for non-stationary systems but also can reduce the computational complexity by reducing the number of confounding variables. Finally, through the Tennessee Eastman chemical process (TEP) as an example to verify the effectiveness of the proposed method.