A machine learning methodology for probabilistic risk assessment of process operations: A case of subsea gas pipeline leak accidents

Xinhong Li,Jingwen Wang,Guoming Chen

DOI: https://doi.org/10.1016/j.psep.2022.04.029

IF: 7.8

2022-04-01

Process Safety and Environmental Protection

Abstract:Subsea gas pipeline leak may cause the catastrophic consequences, e.g., offshore fire and explosion, and the overturning of floating offshore structures. Efficient risk assessment is critical to prevent the unexpected offshore accident due to subsea gas pipeline leak. The modern technology advancement urges the need for developing new risk assessment approach in the case that a large volume of data becomes available. A new methodology comprising of BRANN implemented with BN is proposed for dynamic risk assessment of process operations, which can capture the nonlinear and the non-sequential features of accident escalation. BN is used to model the accident scenario from causations to consequences considering the conditional dependencies among accident contributory factors. Subsequently, BN model as an informative base is mapped into BRANN model. The data-driven nature makes it better to capture the uncertainty among the cause-consequence relationships. The practicability of the methodology is illustrated by a case study of subsea gas pipeline leak accident. It is observed that BRANN model is superior in the network performance and prediction accuracy. This methodology can help to perform more effectively real-time risk assessment of process operations.

engineering, environmental, chemical