Cost–Consequence Predictive Models for Hazardous Liquid and Gas Transmission-Gathering Corroded Pipelines

In the United States, over 4.8 million km (3 million miles) of pipelines form a critical network in transportation of oil and gas. The failure of these pipelines can result in severe environmental, economic, and community consequences. Quantifying these consequences is of importance for effective risk management and maintenance planning of oil and gas pipelines. This study utilizes incident data reported by the US Pipeline and Hazardous Materials Safety Administration (PHMSA) to develop predictive models for financial consequences of failures in hazardous liquid (HL) and gas transmission-gathering (GTG) pipelines due to corrosion. Multivariable regression with all-possible-subset model selection and symbolic multigene regression (SMGR) based on genetic programming (GP) are employed to develop prediction models, using pipeline and incident features as independent variables. The models developed based on regression and GP are compared in terms of model complexity and prediction performance. The sensitivity analysis shows the impact of key features on the consequence prediction, and the findings also highlight the discrepancies between financial consequences used in the literature and the failure costs predicted by data-driven models proposed in this model. Additionally, the impact of variation in the cost consequence of pipelines failures on total expected life-cycle cost and optimal maintenance planning of pipelines is shown through a case study.