Tuning the structure and parameters of a neural network using an orthogonal simulated annealing algorithm

In this paper, an orthogonal simulated annealing algorithm (OSA) is applied to get an optimal network structure and parameters of a feedforward neural network at the same time. An orthogonal experimental design which based on OSA could efficiently generate large good candidate solutions by using a few computing cost. High performance of OSA-based method can be shown to efficiently obtain more accurate solution in prediction of the sunspot numbers problem, compare with other exited methods.