IDENTIFICATION OF GASES WITH CLASSICAL PATTERN-RECOGNITION METHODS AND ARTIFICIAL NEURAL NETWORKS

The paper describes the identification of gases with statistical methods and neural networks. It is shown that there is an optimal standardization of measurement data with regard to prediction accuracy. An example with two sensors and two gases is discussed and the differences of the methods worked out. It is shown in which case neural networks have an advantage over statistical methods. Finally, results of data evaluations with discriminant analyses and neural networks are presented.