System identification using evolutionary Markov chain Monte Carlo

System identification involves determination of the functional structure of a target system that underlies the observed data. In this paper, we presen t a probabilistic evolutionary method that optimizes system architectures f or the identification of unknown target systems. The method is distinguishe d from existing evolutionary algorithms (EAs) in that the individuals are g enerated from a probability distribution as in Markov chain Monte Carlo (MC MC). It is also distinguished from conventional MCMC methods in that the se arch is population-based as in standard evolutionary algorithms. The effect iveness of this hybrid of evolutionary computation and MCMC is tested on a practical problem, i.e., evolving neural net architectures for the identifi cation of nonlinear dynamic systems. Experimental evidence supports that ev olutionary MCMC (or eMCMC) exploits the efficiency of simple evolutionary a lgorithms while maintaining the robustness of MCMC methods and outperforms either approach used alone. (C) 2001 Elsevier Science B.V. All rights reser ved.