Construction of a model for measurement control over liquefied petroleum gas components based on a multilayer perceptron

This study's object is the process of measuring control over liquefied petroleum gas components using a multilayer perceptron. The problem considered is insufficient efficiency of existing methods for measuring control over liquefied petroleum gas components. It can be partially solved by remote measurement of components of liquefied petroleum gas and processing of the received data and, accordingly, control by a neural network. However, another issue arises, associated with the complexity of using neural networks in combination with peripheral devices, in particular, means, sensors, gauges, etc., and the need for significant computing power. This paper reports a model for measuring control over liquefied petroleum gas components, which takes into account its physical characteristics, using a multilayer perceptron, which provides communication with gas measurement devices. The mechanism for achieving these results involves training the model based on performance indicators derived from input data, taking into account the formed features. High generalization ability and efficiency are illustrated by the coefficient of determination, which is 0,845. High accuracy is illustrated by the low overall average value of the mean absolute error, which is 1,1%. That was made possible by the distinctive features of the proposed solution, namely the optimized architecture of the model in accordance with the object of study and its input features. These features are the areas of the light streaks, their logarithmic ratios, temperature, the sum and difference of densities of the components of liquefied petroleum gas. The results can be applied practically to problems involving liquefied gas composition analysis, especially at gas filling stations, oil and gas processing plants, gas storage facilities, and similar sites