Using wavelet neural networks for the optimal design of electromagnetic devices

A feedforward neural network based on the wavelet transform which can be applied to the approximation of complex nonlinear functions is discussed. A wavelet neural network can establish an exact model through a self-adaptive procedure by learning input/output maps from the training sets which are generated by finite element analysis. The structure of the network can be definitely developed, and the learning speed is increased. We have applied it to the optimization design of an AC vacuum contactor with a DC exciting electrical circuit and obtained a satisfactory scheme.