A unified causation prediction model for aboveground onshore oil and refined product pipeline incidents using artificial neural network

Abstract

Aboveground onshore oil and refined product pipeline incidents pose significant hazards to people, property, and environment. Therefore, several data-driven models have been developed to predict their causes and take appropriate troubleshooting decisions. However, these approaches utilize only a few attributes despite the presence of hundreds of attributes in the incident databases. Additionally, existing works focus mainly on sub-cause prediction models and not on cause prediction models. Since the cause of an incident is required to select an appropriate sub-cause prediction model, existing sub-cause prediction models cannot be utilized directly for causation prediction of an incident. To handle these two limitations, this work proposes a unified cause prediction model for aboveground onshore oil and refined product pipeline incidents by leveraging a rich incident database. Specifically, 108 attributes are selected from incident data collected by Pipeline Hazardous Material Safety Administration (PHMSA) for 2010–2019 based on their significance to incidents using domain knowledge. Using selected attributes, first, an artificial neural network (ANN) model is developed to predict the cause of an incident. Secondly, for each cause, another ANN model is developed to predict the sub-cause of the incident. The integration of cause and sub-cause prediction models enables the efficient causation prediction of incidents.