Construction of a model for measuring liquefied gas volume based on an artificial neural network

Abstract

This study investigates the process of quantifying liquefied gas volume using an artificial neural net-work. The task addressed relates to the insufficient efficiency of existing methods for measuring liquefied gas volume. It can be partially solved by measuring the parameters of liquefied gas in cylinders remot ely and by processing the data with an artificial neural net-work to quantify its volume. However, there is another issue associated with the complexity of using artificial neural networks in combination with correspond-ing peripherals, in particular devices, means, sen-sors, gauges, etc., and the need for significant comput-ing power.This paper suggests a model for measuring lique-fied gas volume, which takes into account its physical characteristics, based on an artificial neural network that provides communication with gas measurement devices. The mechanism behind such result involves training the model based on performance indica-tors derived from input data, taking into account the formed features. High generalization ability and efficiency are illus-trated by the coefficient of determination, which equals 0.999245. High accuracy is illustrated by the overall low average value of a mean absolute error, which equals 1%. That was made possible by the dis-tinctive features of the proposed solution, namely the optimized model architecture in accordance with the object of study and its input features. These features are the signal from a photodetector, which character-izes the level of liquefied gas, the angles of the cylinder in the vertical plane, as well as in the horizontal plane.The results could be applied to tasks involving the measurement of liquefied gas volume, especially at oil and gas processing plants, gas filling stations, gas stor-age facilities, etc.Keywords: liquefied gas, artificial neural network, mean absolute error, coefficient of determination.