CHOICE OF OPTIMUM MODEL PARAMETERS IN ARTIFICIAL NEURAL NETWORKS AND APPLICATION TO X-RAY-FLUORESCENCE ANALYSIS

The model parameters in artificial neural networks have a great influe nce on the training speed. It can be increased after choosing the opti mum parameters, which was performed by a stepping technique. The train ing speed using the method is usually faster than that when adopting r andom or empirical parameters. An artificial neural network model was used in multivariate matrix calibration and compared with cross-valida tion and partial least-squares methods, which mere combined with the f undamental-parameters in x-ray fluorescence analysis. The results show that the artificial neural network model produced the highest accurac y.