A study on radial basis function neural network size reduction for quantitative identification of individual gas concentrations in their gas mixtures

In this study, the multilayer neural networks (MLNNs) with sigmoid hidden layers and radial basis function neural networks (RBFNNs) were compared for quantitative identification of individual gas concentrations in their gas mixtures (trichloroethylene and n-hexane), and a method to reduce the RBFNN size for quantitative analysis of gas mixtures was proposed. For this purpose, three MLNNs and three RBFNNs structures were applied. A data set consisted of the steady state sensor responses from the quartz crystal microbalance (QCM) type sensors was used for the training of the first MLNN and RBFNN. The other MLNNs and RBFNNs were trained using two different reduced training data. The components in the binary mixture were quantified applying the sensor responses from the QCM sensor array as inputs to the MLNN and radial basis neural networks. The performances of the neural networks were compared and discussed based on the experimental results.