Applies of Neural Networks to Identify Gases Based on Electronic Nose

Intensive research and fast developments in electronic nose (EN) technologies provide the users with a wide spectrum of sensors and systems for their applications. Back-propagation neural network (BP), radial basis function neural network (RBF), and self-organization mapping networks (SOM) were applied to identify three gases by electronic nose gas sensors (CO, SO₂, and NO₂) qualitatively. Three training algorithms, gradient descent (traingd), gradient descent with momentum of variable learning rate (traingdx) and Levenberg-Marquardt (trainlm) algorithm, were applied for training. The results show the first two algorithms are too slow for practical problems. Training speed of trainlm is faster more. The RBF networks provide a simple and robust method. The sampling gases were clearly classified with few errors. The RBF networks train faster than the BP networks do, while exhibiting none of back-propagation´s training pathologies such as paralysis of local minima problems. The SOM networks can classify accurately and generalization capability is far superior. While recognized patterns are non-rectangular shape and size, the performance is poor.