Odour quantification by a sensor array: An application to landfill gas odours from two different municipal waste treatment works

The possibility of quantifying the landfill gas (LFG) odour in terms of odour-units per cubic meter (ou/m³) using a tin oxide sensor array is investigated. The objective is to determine the most appropriate neural machines (MLP networks, RBF networks) model to perform the odour concentration approximation and evaluate the influence of multiple biogas sources modelling on the approximation quality. The structural risk minimization principle is used instead of the usual empirical risk minimization principle in the training algorithm of the neural machines. Multilayer perceptrons (MLP) networks prove to minimize best the error on the prediction of odour concentration of unknown data. The data is constituted of LFG odour samples from two municipal waste treatment works presenting different concentrations of odorous compounds. It is shown that the quality of the LFG odour approximation is in the present case influenced directly by the size of the training data set. The use of data coming from two different sources is not detrimental to the quality of the approximation.