Global Optimization by Annealing Type of Evolutionary Programming and Its Application to System Identification Problem Using Neural Networks
The global optimization procedures which simulate the natural evolution are known as Evolutionary Programming or Simulated Evolution. We utilize several concept from natural evolution and construct the global optimization procedure. The features of the proposed method are summarized as follows :
1) Introduction of the concept of temperature to the reproduction process
2) Ajustment of the temperature and metric ; in early stages the generated points can be moved drastically in order to escape from the local minimum and the convergence to the global solution in the final stage should be satisfied.
3) The reproduction and selection processes are simple.
The numerical examples including Shekel function type problem are solved and the proposed method hardly misses the global solutions. The identification problem of a nonlinear system by neural networks is also solved successfully. The proposed method is very promising and can be applied to many types of real problems.