Optimum grouping in a modified genetic algorithm for discrete-time, non-linear system identification

• Main Author: Abd Samad, Md Fahmi
• Format: Article
• Language: English; English
• Published: SAGE 2007
• Subjects: TA Engineering (General). Civil engineering (General)
• Online Access: http://eprints.utem.edu.my/id/eprint/3972/1/03_975-990_JSCE362.pdf; http://eprints.utem.edu.my/id/eprint/3972/2/impact_factor_2007.pdf; http://eprints.utem.edu.my/id/eprint/3972/; http://www.uk.sagepub.com/journals/Journal202033
• Institution: Universiti Teknikal Malaysia Melaka

The genetic algorithm approach is widely recognized as an effective and flexible optimization method for system identification. The flexibility of a genetic algorithm allows various strategies to be applied to it. One of the strategies applied is the modified genetic algorithm which relies on, among other things, the separation of the population into groups where each group undergoes mutual recombination operations. The strategy has been shown to be better than the simple genetic algorithm and conventional statistical method, but it contains inadequate justification of how the separation is made. The usage of objective function values for separation of groups does not carry much flexibility and is not suitable since different time-dependent data have different levels of equilibrium and thus different ranges of objective function values. This paper investigates the optimum grouping of chromosomes by fixed group ratios, enabling more efficient identification of dynamic systems using a NARX (Non-linear AutoRegressive with eXogenous input) model. Several simulated systems and real-world timedependent data are used in the investigation. Comparisons based on widely used optimization performance indicators along with outcomes from other research are used. The issue of model parsimony is also addressed, and the model is validated using correlation tests. The study reveals that, when recombination and mutation are used for different groups, equal composition of both groups produces a better result in terms of accuracy, parsimony, speed, and consistency.