关键词: 随机森林, 故障诊断, 数据不可靠, 机器学习, TE过程

Abstract: The monitoring data of chemical process has unreliable problems such as data missing, drifting and stuck, which may seriously affect the accuracy of fault diagnosis. To solve the problem, a fault diagnosis method which is based on random forest (RF) was proposed. The RF classifier was trained and tuned by using the training set to obtain the optimal RF classifier model. The numbers of decision trees and the random attributes were determined. Finally, the test set with unreliable data was input into the RF classifier model to diagnose the faults using the advantage of strong anti-interference ability of RF. Tennessee-Eastman (TE) process was used to test the performance of the proposed RF based fault diagnosis method. Testing results show that compared to the Back Propagation Neural Network (BPNN), Radial Basis Function Neural Network (RBFNN) and Deep Belief Network (DBN), the RF-based fault diagnosis method can detect and identify fault types more effectively with unreliable data. It is shown that the proposed method is superior in the application of practical industrial environment.

Key words: random forest, fault diagnosis, unreliable data, machine learning, TE process