Optimum design using radial basis function network and adaptive range genetic algorithms (3rd report, usage of data generation by using self-organizing maps)

Genetic Algorithms have been studied widely to report their effectiveness in a variety of fields. As optimizer, they give us much privilege in considering types of design variables for multi-peaked problems. However, there exist some shortcomings, such as treatment of continuous variables, formulation of fitness function and the number of function calls. In the previous studies, we have been developed Adaptive Range GAs for continuous variable problem, by using RBF (Radial Basis Function) networks as approximation tools in solving large scale constraint optimization problems. From these results, we have shown the effectiveness to obtain good results. However, it is very difficult to give data within feasible region only by using random numbers for RBF and almost 80% of them are consumed in vain. In order to raise their ratio, we introduce SOM (Self-Organizing Map) as classification tools to choose data within feasible region. In this study, we will show the effectiveness of the proposed method based upon a famous benchmark test problem.