Affiliations: Array and Information Processing Laboratory, College of Computer and Information Engineering, Hohai University, Nanjing, Jiangsu, P.R. of China.
Correspondence:
[*]
Corresponding author. Shuang Wei, Array and Information Processing Laboratory, College of Computer and Information Engineering, Hohai University, Nanjing, Jiangsu 210098, P.R. of China. Tel.: +86 0255 8099106; Fax: +86 0255 8099124; E-mail: wishuang@gmail.com.
Abstract: This paper presents a new multiobjective genetic programming (MOGP) approach, to realize an all-in-one automatic nonlinear system design (NSD). The nonlinear system design is here modeled as a multiobjective optimization problem (MOP) to solve parameter estimation, structure optimization and feature selection simultaneously. The novel MOGP method is then proposed to rank individuals according to the ‘compromise distance’ between them, which has the benefit of combining decision making for NSD with the optimization process to get the final compromise solution in a single process. The effectiveness of the proposed learning approach for nonlinear system design is verified through experiments on the classical nonlinear autoregressive with extra inputs (NARX) system by comparison with classical aggregating method and a Pareto-based method for MOP. Finally, experimental results demonstrate the proposed approach is available to explore the unknown structure of nonlinear systems as well as the features and parameters with high accuracy and efficiency.
