A Non-Greedy Algorithm Based L1-Norm LDA Method for Fault Detection

Min Wang,Maoyin Chen,Jingxin Zhang,Donghua Zhou

DOI: https://doi.org/10.1109/safeprocess45799.2019.9213401

2019-01-01

Abstract:$L_{1}$ -norm based linear discriminant analysis ( $L_{1}$ -LDA) has attracted much attention. Compared with traditional $L_{2}-$ norm based linear discriminant analysis ( $L_{2}$ -LDA), $L_{1}$ -LDA has better robustness against outliers and noise. However, for most forms of $L_{1}$ -LDA technique, optimal projection vectors are calculated one by one using greedy strategy. Compared with $L_{2}-$ LDA, the objective function is to maximize the ratio of $L_{1}$ norm instead of $L_{2}$ norm. It cannot be guaranteed that the obtained optimal projection matrix is globally optimal. In order to simultaneously consider the robustness of the anomaly data and the global optimal solution, this paper proposes a non-greedy algorithm based $L_{1}$ -norm LDA method for fault detection. And the effectiveness of the proposed method is illustrated by a numerical example and the Tennessee Eastman process.