Exploring the temporal structure of time series data for hazardous liquid pipeline incidents based on complex network theory

The pipelines that transport hazardous liquids (e.g., petroleum and petroleum products) across cities or countries are critical to the energy-supply infrastructure, and they are crucial for the reliable and secure operation of a city. Estimating the occurrence rate of serious pipeline accidents with sparse data is a challenging problem in pipeline safety management because serious hazardous liquid pipeline accidents are caused by a particular multidimensional sequence of events. In this paper, complex network theory was employed to detect the temporal structure of pipeline incidents and reveal the nonlinear connections between major accidents and their precursors. A database of hazardous liquid pipeline incidents in the US between 2010 and 2018 collected by the Pipeline Hazardous Material Safety Administration (PHMSA) of the US Department of Transportation was transformed into a complex network via the visibility graph algorithm. The temporal structure of the pipeline incident time series for different years and different companies was explored by applying complex network analysis. The results show the scale-free property and small-world topology of the constructed networks and provide the rationale for applying the hierarchical Bayesian model to predict the occurrence rate of major accidents in a pipeline system when there is sparse data. The benefits of using the hierarchical Bayesian model for estimating the occurrence rate were illustrated by comparing it with three different methods. Furthermore, posterior predictive checks were performed to validate whether the results of the hierarchical Bayesian model are consistent with the real data. The result indicates that it is reasonable to apply the hierarchical Bayesian model to estimate the occurrence rate of serious pipeline accidents when there is sparse data. Finally, practical applications of the temporal structure of the incidents were proposed for improved pipeline safety management. Our results provide the underlying new insights needed to enhance quantitative analyses of pipeline incidents.