The Comparation of Back Propagation Method and Kohonen Method for Gas Identification

The identification process of gas flavor is conducted by using the output of gas sensor system to recognize a variety of gas flavor. The identification and analysis process of this system is processed by using an artificial neural network approaches those are back propagation and Kohonen method. According of the experiment’s result, the best parameter for back propagation network is the momentum constant (α) = 0.7, the constant of the sigmoid function (β) = 4.5, constant learning (η) = 0.9, and the constant of convergence (ε) = 0001, convergence is achieved more or less in the 19 500 iterations (± 16 seconds). Meanwhile, the best classification for Kohonen network is for the output of 8 knots with an average of 80.7% uniformity (for a maximum of 500 times iteration, approximately ± 3 seconds). Thus, the best network to classify the signal pattern of gas flavor is back propagation network for the parameters (α) = 0.7, a constant sigmoid function (β) = 4.5, a constant learning (η) = 0.9, and a constant convergence (ε) = 0001.