Deep ensemble forests for industrial fault classification. (30th December 2019)

• Record Type: Journal Article
• Authors: Liu, Yue; Ge, Zhiqiang
• Is Part Of: IFAC journal of systems and control. Volume 10(2019)
• Publication Date: 2019-12-30
• Subjects: Ensemble learning -- Deep ensemble forests -- Feature extraction and selection -- XGBoost -- Random forests -- Fault classification; Automatic control -- Periodicals; Relay control systems -- Periodicals; Embedded computer systems -- Periodicals; Feedback control systems -- Periodicals; Artificial intelligence -- Periodicals; Artificial intelligence; Automatic control; Embedded computer systems; Feedback control systems; Relay control systems; Electronic journals; Periodicals; 629.89
• DOI: 10.1016/j.ifacsc.2019.100071
• Languages: English
• ISSNs: 2468-6018

Abstract

Abstract: Taking into account different data characteristics and assumptions, it is hard to define a single classifier that can achieve desired fault classification performance under different circumstances. To tackle this problem, a feature extraction and feature selection based deep ensemble forests model is proposed in this paper, which uses XGBoost, random forests and extremely randomized trees as basic forests in each layer to improve the diversity and the accuracy. In this model, the input of each layer is the concatenation of class probability vector produced by the previous layer and the selected feature vector. The cascading process will be automatically terminated while the performance of the layer no longer increases. The feature importance obtained at first layer is used for feature selection, which is able to make the model more efficient and avoid overfitting. Furthermore, a weighted probability fusion method is employed at the last layer for the final decision. A case study on Tennessee Eastman (TE) benchmark process is conducted and gives a comparison between proposed method and conventional methods.