Infrastructures are multi-agent systems. In other words, they consist of subsystems that are more or less autonomous, but related to each other in a hierarchical, coordinating or cooperative way. Each of the subsystems operates according to its own objectives and interests. Each of them, as a result, follows its own operational regime. But how does the behaviour of each subsystem affect the behaviour of the system as a whole?
To answer that question, researchers study not just the individual subsystems, but also the connections between them. This allows them to predict and understand the phenomenon of emergence: system behaviour that cannot be predicted by only looking at the lower components.
One of these researchers is Koen van Dam. As part of his research project, he is currently working in Singapore, where he studies the operations of a large oil refinery. “An oil refinery is a classical example of a multi-agent system,” says Van Dam. “Firstly, there are the company’s different departments, such as purchase, sales and operations. Then there are external agents, such as consumers, various selling agents and shipping companies. By regarding these agents as autonomous subsystems and looking at their interrelationships, we can describe the system as a whole.”
Van Dam and his colleagues do this through advanced modelling methods. They are designing so-called ontologies: formal concepts in ‘computer language’ that can be used to describe a system. Agents can use these ontologies to communicate with each other.
“It comes down to a standardised way of communication,” he says, “based on terms that are clearly defined.” When running his model, Van Dam can detect and study emergent behaviour of the system. He aims to describe the general patterns: which strategies work and which don’t? And why?
“The nice thing about a model,” Van Dam states, “is that it allows you to play with the different variables. You can build a system from the ground and see how your decisions affect the overall system.” Eventually, these models may be used as decision support tools for policy makers and network managers. In the process, scientists hope to identify how such models can be developed and what the appropriate modelling paradigms are.
