CO2 Concentration Prediction in Natural Gas: A Machine Learning Approach for Optimizing Operational Efficiency and Production Quality

Abstract Natural gas as a cleaner energy source emits less carbon dioxide (CO₂) when burned than other fossil fuels. During production, natural gas contains several impurities, and a higher CO₂ concentration in natural gas reduces its heating value, which causes potential pipeline corrosion for the gas transportation system. Conventional methods for predicting the concentration of CO₂ in natural gas, like gas chromatography, the equation of state, and empirical correlation, are time-consuming and may become less accurate when handling non-linearity with different gas composition variables. The dataset explored in this research was obtained from a gas production company in the Niger Delta region of Nigeria. Four machine learning models, including support vector regressor, AdaBoost regressor, gradient boosting regressor, and random forest regressor, were developed to predict the concentration of CO₂ in natural gas with six features: methane, ethane, butane, propane, density, and heating value. Data preprocessing techniques were carried out to handle duplicated and missing values. An exploratory data analysis (EDA) was implemented using a heatmap to understand the relationship between selected features and the target variable. With an R?2 accuracy of 0.997 and a mean absolute percentage error of 3.55%, the random forest model performed better than other machine learning models. A user-friendly Gradio interface was used to deploy the developed model for easy accessibility by users. The high accuracy of the model demonstrates its potential for real-time CO₂ concentration monitoring, enabling more efficient gas processing, reducing environmental impact, and ensuring compliance with pipeline specifications.