Stochastic nonlinear system identification using multi-objective multi-population parallel genetic programming

To realize simultaneous identification of both structures and parameters of stochastic nonlinear systems, multi-population parallel genetic programming (GP) was employed. Object systems were represented by nonlinear autoregressive with exogenous inputs (NARX) and nonlinear autoregressive moving average with exogenous inputs (NARMAX) polynomial models, multi-objective fitness definition was used to restrict sizes of individuals during the evolution. For all examples, multi-population parallel GP found accurate models for object systems, simultaneously identified structures and parameters. In comparison with traditional single-population GP, multi-population GP showed a more competitive performance in avoiding premature convergence, and was much more efficient in searching for good models for object systems. From identification results, it can be concluded that multi-population parallel GP is good at handling complex stochastic nonlinear system identification problems and is superior to other existing identification methods.