Detection of magnetic body using artificial neural network with modified simulated annealing

Abstract

An artificial neural network is applied to the inverse electromagnetic fields problem. In the process of the training the network, it is suggested that the simulated annealing algorithm be used to smooth the output errors before the network is trained with the error back-propagation algorithm. And a general way of defining the control parameters of simulated annealing is presented. As numerical example, the artificial neural network with the suggested training algorithm is applied to the detection of the magnetic body in magnetic field. It is shown, through the numerical test, that the artificial neural network is very useful for the inverse electromagnetic field problems, especially in real-time system and the artificial neural network trained with the suggested training algorithm gives much less maximum errors than that trained with the error back-propagation algorithm only. >