Using artificial neural networks for solving the problem of analysis of the composition of gas mixtures

The problem formulation of natural gas composition analysis is described. The statistical method is proposed to use for solving the problem of gas composition analysis. The main stages of the statistical model development for the natural gas composition analysis are described. The results of the correlation analysis for the selection of input and output parameters for the statistical model are presented. The main statistical models that used for solving of the problem of gas mixtures composition analysis are shown. The artificial neural networks and Levenberg – Marquardt algorithm are used within the study. The description of the Levenberg – Marquardt training algorithm is given taking into account possible modifications of the algorithm. The architecture of the proposed neural network is described. The gas mixtures ranges that used in the training and test samples are given. The accuracy characteristics of the proposed model are given. It was concluded that the chosen neural network model architecture is adequate, based on the calculated accuracy characteristics of the model. The results of predicting the gas mixtures composition by measurements of gas physical parameters are shown. Further research directions in the development of the proposed method for analyzing the natural gas composition are given.