Network and agent-based modeling to identify petrochemical supply chain vulnerabilities

Image of hemical plant at sunset
Diagram of petrochemical supply chain entities and relationshipsThe chemical sector depends on both several critical infrastructures (principally Petroleum, Energy, and Transportation) and a complex, globally distributed, multistage processing chain for their successful operations. Understanding how these components work together under normal and disrupted conditions is critical for achieving asset prioritization, consequence assessment, and policy guidance goals.

When disruption occurs, the CASoS-based Loki model identifies potentially affected chemicals and technologies by representing the chemical industry as a network of chemical dependencies. This modeling, simulation, and analysis capability can be used to assess sector vulnerabilities, its interdependencies with other critical infrastructures, its potential impacts from disruptive events (such as manmade and natural disasters), and its overall economic resilience.

The Loki toolkit embodies a generalized network and agent-based approach and contains a set of components that can be selected, specialized, and combined to create models of diverse systems including power systems, pipelines, social networks, and financial systems, as well as interactions across these different networks/systems. Loki has been applied to generic congestive cascade, power grids, payment systems, social simulation, and infectious diseases as well as petrochemical and natural gas systems.

  • LOKI-Network algorithms and techniques have been used to analyze the petrochemical subsector to predict the nonlinear impact of the loss of typical and atypical production capacities on overall systemic throughput.
  • The high-level view in this idealization of chemical supply chains was also used to identify problematic areas. For example, a network analysis reveals that propylene and styrene are connected to many other chemical products; such interconnectedness merits special attention from other higher-fidelity modeling approaches.