Application of adaptive RBF network for odour classification under drift effect using conducting polymer sensor array

Most of chemical sensors do suffering aging or poisoning problems over the long period of time. This drift causes electronic nose to predict odours difficulty after some period of time. We applied an adaptive Gaussian radial basis function (RBF) network based on the Stochastic Gradient (SG) method to perform for drift compensation. The centers and widths of RBF Network were finely tuned by SG method and weights between hidden and output layer were calculated by Singular value decomposition (SVD) to minimize prediction error in iterative learning stage. The adaptive RBF network has shown good odour prediction performance after some period of time, even if the sensors suffer from aging and poisoning problems. It was confirmed by experimental trails.