A Novel Local Selective Ensemble-based AdaBoost Method for Fault Detection of Industrial Process

For the sake of guaranteeing the security of complex industrial system, it is important to accurately and efficiently detect the faults. AdaBoost algorithm is an effective fault detection method. It can generate a large number of weak classifiers in iterations and combine many of these weak classifiers into the strong classifier to solve the classification problem for fault detection. For the traditional AdaBoost, several of these poor weak classifiers are often ignored and not fully used. However, the weak classifiers with poor performance may store the significant information and pay more attention to the difficult samples. To solve these problems, we propose a local selective ensemble-based AdaBoost (AdaBoost-LSE) in this article. Firstly, error feedback ELM (EFELM) is introduced to establish the basic weak classifier. Through the iteration of AdaBoost, these weak classifiers based on EFELM are generated. Secondly, these weak classifiers are divided into good weak classifiers and bad weak classifiers based on the classification accuracy. The poor weak classifiers with good performance are selected by calculating the classification accuracy for the targeted samples. Thirdly, the strong classifier of AdaBoost-LSE is constructed by integrating the original good weak classifiers and some of these poor weak classifiers with good performance. To verify the efficiency of AdaBoost-LSE, the Tennessee Eastman (TE) simulation process is used. The experimental results reveal that the proposed AdaBoost-LSE can greatly improve the accuracy of fault detection.