Predictive Analytics Model for Natural Gas Transportation Consumption

Abstract In the current natural gas transportation, operational scheduling heavily relies on daily volume nominations submitted by shippers. However, inaccuracies in these nominations pose significant challenges, leading to imprecise planning and complicating transportation operations. With the advancement of the Fourth Industrial Revolution, driven by automation and artificial intelligence (AI), revolutions are expected across various industries, promising significant improvements in efficiency, safety, and quality. This study focuses on predicting daily consumption in a natural gas distribution network, utilizing statistical techniques such as time series analysis and regression models. Additionally, advanced Machine Learning algorithms, such as XGBoost, are incorporated to refine predictive capabilities. The prediction algorithms were developed using the Python programming language. Historical data spanning three years underwent rigorous preprocessing before model training and testing. Key metrics such as Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) were employed to assess the model’s performance, showcasing its superior accuracy in predicting consumption volumes at delivery points compared to the initial nominations received from shippers. The algorithm encompasses operational data point analysis, outlier handling, hyperparameter optimization, and model tuning. This is followed by integration into the operational engineering and logistics workflow for ongoing operational evaluation. These findings underscore the potential of advanced analytics and machine learning to tackle operational challenges and enhance efficiency and reliability in natural gas transportation.