Identifying Stochastic Nonlinear Dynamic Systems Using Multi-objective Hierarchical Fair Competition Parallel Genetic Programming
Xiao-lei Yuan and Yan Bai
A parallel evolutionary algorithm named hierarchical fair competition genetic programming (HFC-GP) was employed to identify stochastic nonlinear dynamic systems. Nonlinear autoregressive with exogenous inputs (NARX) and nonlinear autoregressive moving average with exogenous inputs (NARMAX) polynomial models were used to represent object systems. Multi-objective fitness was used to restrict individual structure sizes during the run. HFC-GP outperformed single-population GP and traditional multi-population GP in combating premature convergence. For all examples, good results were achieved with simultaneous and accurate identification of both structures and parameters. It can be concluded that HFC-GP is very effective in combating premature convergence and is superior to other exiting identification methods.
Keywords: Nonlinear dynamic system identification, Stochastic system identification, NARX, NARMAX, HFC-GP, multi-objective evolution.