The effect of ANN hidden layer neurons on boundary identification for constrained optimization

Summary form only given. The process of constrained optimization involves first finding a feasible region which satisfies all design criteria and then optimizing the design goal while remaining in this feasible region. The constraint boundaries, which are the borders between feasible and infeasible regions, were identified using the pattern classification capability of backpropagation networks. In addition, various numbers of hidden layer neurons were chosen to study their effects on boundary approximation. The results showed that the identified boundaries did evolve from a crude approximation to a shape that is close to true boundaries. The increase in the number of hidden layer neurons did not improve the results of the training significantly. These preliminary results indicated that ANN (artificial neural network) pattern classification training is robust against the choice of hidden layer neurons and the procedure of constraint boundary identification can be incorporated into constrained optimization problems