Modeling of Natural Gas Consumption Volumes in China Aided by Machine Learning Methods

Abstract

The paper applies machine learning methods to make projections of natural gas consumption volumes in China. It is critical for Russia as a major supplier of natural gas to China to have a reasonable estimate of possible volumes of exports. This contributes to the proper allocation of available raw materials, reduces the cost of excess gas storage, and also facilitates long-term planning for future trade. These aspects are critical for sustaining Russia's economic security and developing international economic relations. It is possible to estimate possible exports based on the projected volumes of natural gas consumption in China. This study uses machine learning methods, which are considered a promising data analysis tool, to model such consumption. We used multiple models for their benchmark comparison. Ridge regression was used as a linear model, whereas random forest and gradient boosting served as nonlinear models. The simulations performed proved gradient boosting to be the best choice. The study revealed the decisive role of socio-demographic factors, such as the population and the urban area size. The most significant factors were the total population, gas reserves, urban area size, number of passenger cars, and the population in urban agglomerations with over 1 million inhabitants.