Building Machine Learning of a Predictive Model of Natural Gas Consumption Based on the Variability of External Factors

This paper is devoted to the development of a predictive model of natural gas consumption using machine learning algorithms. The development involves the creation of a short- and long-term forecast of gas consumption. The urgency of solving this problem at the present time is due to the growing need to increase efficiency and stabilize both the global gas industry and the Russian one. The development of a predictive model will make it possible to predict long-term and short-term changes in gas consumption, on the basis of which the most energy-efficient mode of operation of the gas transmission system is selected, thereby reducing operating costs or, in the case of a long-term forecast, effectively plan capital expenditures. The purpose of this work is to increase the efficiency of the gas transportation process through the effective implementation of a machine learning model in order to predict gas consumption in the short and long term. To achieve the stated goal, the following tasks were set: • Conduct a comprehensive analysis of the structure of natural gas consumption in the target city to identify key metrics that affect its consumption. To assess the degree of influence and dependence of these metrics. • Prepare and clean the dataset necessary for building the model, ensuring its reliability and completeness. • Develop long-term and short-term predictive models that ensure sufficient accuracy of predictions. To analyze the obtained long-term and short-term forecasts, to evaluate the effectiveness of the model and its impact on planning and management of gas resources • To calculate the economic effect of using this development. The object of research in this work is the process of natural gas consumption in the context of a unified gas supply system. The subject of the research is natural gas consumption metrics, meteorological data, demographic and economic indicators for the period from 2013 to 2023. The methods used in this work for the purposes of the study are: a review of scientific publications, books, articles, online resources on the topic of the study and their analysis, data collection on such implemented projects and their analysis, comparison. The scientific novelty of this work lies in the originality of the approach of using such a tool as machine learning to solve the problem of forecasting gas consumption in the context of Russian gas supply. This task has not been solved using machine learning methods in Russia before, and an extremely limited number of studies are devoted to the analysis of gas consumption in cities with significant consumption. Thus, the work represents an innovative contribution to the field of energy management, providing a new approach to forecasting and optimizing natural gas consumption on main gas pipelines, which is an urgent and promising direction in the context of modern challenges of energy efficiency and automation.