Linear programming svm-arma with application in engine system identification

Abstract—As an emerging non-parametric modeling technique, the methodology of support vector regression blazed a new trail in identifying complex nonlinear systems with superior gener-alization capability and sparsity. Nevertheless, the conventional quadratic programming support vector regression can easily lead to representation redundancy and expensive computational cost. In this paper, by using the norm minimization and taking account of the different characteristics of autoregression (AR) and the moving average (MA), an innovative nonlinear dynamical system identification approach, linear programming SVM-ARMA, is developed to enhance flexibility and secure model sparsity in identifying nonlinear dynamical systems. To demonstrate the potential and practicality of the proposed approach, the proposed strategy is applied to identify a represen-tative dynamical engine model. Note to Practitioners—The identification of complex nonlinear engine systems poses challenges on the efficiency, flexibility and generalization capability of the identification algorithms. Inspired by the triumphs of support vector learning methodology in pattern recognition and regression analysis, an innovational nonlinear sys-tems identification algorithm, LP-SVM-ARMA was developed in this paper as a significant stride toward the fusion of modern machine learning essentials into practical identification strategy for industrial systems. In particular, the possible generalization of LP-SVM-ARMA via more complex composite kernel functions is also discussed to meet the diversity of industrial practice. Index Terms—Composite kernel, generalization, linear pro-gramming, model sparsity, nonlinear system identification, support vector regression. I