Surrogate Models for the Compressibility Factor of Natural Gas

The paper presents an example of the so-called surrogate modeling. This is a computer modeling technique where machine learning methods are used to build a fast (surrogate) model that allows you to get a result with acceptable accuracy on data from a complex (physically proven, built on a solution of systems of nonlinear algebraic equations or partial differential equations) and resource-intensive model of an object or process. We develop and analyze the models for calculating the compressibility factor trained on a large amount of data calculated with AGA-8 equation of state. The presented models can be applied to replace the original model when analyzing the development of risk situations and searching for optimal gas transportation modes, in software designed for staff training on computer simulators. The feature and the novelty of the proposed study is not only an analysis of the accuracy of the obtained models for the complex multicomponent composition of natural gas, but also an analysis of the derivatives of the compressibility factor with respect to pressure and temperature, which is important for conducting calculations of nonstationary modes of gas transmission and calculating the sound speed in the gas.