Harnessing Data Analytics and Machine Learning to Forecast Greenhouse Gas Emissions

Abstract

Summary Recently, the attention of all industries is moving toward mitigating climate change through actionable measures, with a focus on CO2, CH4, SOx, NOx, and other gases and particulate matter (PM). The gases together form greenhouse gas (GHG) emissions. The oil and gas industry is actively developing solutions and workflows to reduce emissions across their respective value chains. The first critical step in this is to understand the underlying cause through measurements. Measured data can be analyzed to conduct gap analysis and propose appropriate solutions. In this study, emission data has been used, measured by high-quality sensors during hydraulic fracturing, which is a high-emission intensity operation. We subject these high-frequency measurement data to a comprehensive machine learning (ML) workflow used to create a predictive model based on simple predictors such as temperature and relative humidity measurements. Multiple regression algorithms within supervised learning techniques have been implemented to test their accuracy and show promising results. The implications of this study can potentially enhance operational efficiency and lower the cost of quantifying emissions.