Abstract
By nature, gas distribution is a network system; it does not fit well with traditional pipeline risk models that assume a linear geometry. Distribution system growth is multigenerational and often leads to mixed assets in the same area where transmission pipeline segments are often constructed within shorter time frames and more uniform materials. Slow, sporadic growth leads to varied record availability and quality that might not readily support commercially available risk models. The project described in this paper was initiated to develop predictive methods to prioritize mitigation and replacement activities for distribution networks. Priority is assigned to areas by risk, equipment characteristics and environmental attributes.
In a previous IPC paper, the authors developed a histori-calbased predictive model but applied it to a single city area. This work has been extended to cover the entire province of Saskatchewan. The model relies on logistic regression and a machine learning algorithm to associate the historical failure rate with the asset type, age, pipe material, diameter, pressure, and an array of geographical-dependent attributes such as soil properties and climate events. The output of the model allows integrity engineers to consider predicted failure rates to complement the lagging performance indicators used to develop integrity program planning. This model demonstrates the advantage of using available distribution system records to develop a custom historicalbased predictive model.
Consequence estimates for distribution networks are also described. Distribution leaks are often classified into hazard levels that differentiate operational response. These are assigned based on incident data records and SME input to develop an event tree for the consequence of a distribution leak. This paper summarizes the work performed during the project to calculate distribution asset probability of failure and consequences.
