Non-linear system identification using genetic algorithms

Abstract

In this paper, a genetic-algorithm-based non-linear autoregressive with exogenous inputs system identification (GANARXSI) algorithm was developed to identify non-linear systems and was successfully applied to both non-linear continuous-time and discrete-time systems with acceptable accuracy. This algorithm can achieve robustness and efficiency in identifying complex non-linear systems. The effects of different combinations of operators of genetic algorithms are also studied and an operator of truncate mutation was proposed to improve greatly the convergence rate in identification process. An identified non-linear coupled liquid-level system is used to evaluate the performance of the algorithm. An experimental verification is carried out in a dynamometer engine test cell and the identified engine model during the I/M240 driving cycle is also used to validate the proposed methodology. After testing the algorithm for different complex non-linear systems, the GANARXSI algorithm proved to be a practical technique of identification for non-linear systems.