The advantages and disadvantages of centralization and decentralization are themes commonly discussed in both politics and organization leadership. However, these concepts are also highly relevant within the field of engineering. By studying large-scale networks, such as traffic systems and wind farms, researchers within Large-Scale Systems and Learning have proven global optimality using only local information, generating more efficient as well as more resilient networks with less environmental impact.
Giacomo Como is an associate professor at the Department of Automatic Control.
“It may seem somewhat surprising that local information can give global optimality, but it is a recurring theme within this type of research”, says Giacomo Como.
This is an important finding for traffic networks in particular. “We have reached a situation where the capacity of our infrastructure is very limited. The traffic load is increasing continuously, but there is not enough space or money to meet these demands in terms of newer or wider roads.”
The consequences? People are sitting in queues for hours and hours every day, and, sometimes, massive traffic jams occur. An extreme example is the China national highway 110 traffic jam in 2010, causing congestion of thousands of vehicles for more than 100 kilometers, and lasting for more than ten days.
Smart functions help avoid congestion
“The resilience of a system can be improved using control techniques to design traffic systems with smart functions, avoiding this kind of massive congestion in the future, saving both time and money, as well as reducing pollution”, says Giacomo Como.
These smart functions include variable speed limits, ramp metering (i.e. limiting the number of vehicles that are allowed to enter a freeway at a particular time), dynamic traffic lights, and routing choices such as prescribed turning ratios at junctions (through navigation support in vehicles). Both tolls and incentives have been tested in several parts of the world, with the aim of routing traffic more efficiently.
Dynamic models increase resilience
Giacomo Como and his colleagues have also adopted another significant shift in their approach to large-scale networks. “Static models have long been regarded as the fundamental design paradigm for infrastructure systems, and represent an important area of mathematical programming. But there is increasing awareness that the full potential of large-scale systems can only be properly understood in a dynamical context”, says Giacomo Como, and continues:
“Using dynamic models, we are trying to design control parameters that make traffic networks more resilient, and reduce the effects of shocks that inevitably occur in a system, instead of leading to serious problems.”. Dynamic modeling of large networks is one of the major trends in control theory today, not only in traffic systems, but also, for example, in financial and social systems. The research carried out on resilient network dynamics has received significant international recognition, and, in 2015, Giacomo Como and his colleagues were awarded the G.S. Axelby Award for an Outstanding Paper, for their work on distributed routing in dynamical networks.
“We introduced the concept of ‘the margin of resilience’, which is a way of measuring how resilient different controls are in different types of networks”, says Giacomo Como.
Collaborations around the world
In order to test their dynamic models, Giacomo Como and his colleagues at the Department of Automatic Control are collaborating with research groups in both Central Europe and the USA, where traffic networks are considerably larger, and the traffic more intense than in Sweden. Giacomo Como believes that the area of transportation provides many opportunities for future research.
“Everyone is talking about self-driven smart cars nowadays. Just imagine all the control issues that will emerge in the co-existence between self-driven cars and traditional means of transportation, which will definitely be around for a long time to come. I believe many of our future research challenges will be in the interaction between the two”, says Giacomo Como.
Centralized and Distributed Systems
The traditional control paradigm has been a centralized architecture in which all the information is collected in a single control unit, where it is processed, and the response determined. However, such centralized architectures have proved unsuitable for the control of increasingly complex network systems, as they rely on the assumption that all the information is available, delay-free, in the central control unit. A shock in one part of the network system often spreads into the rest of the system, creating a cascading effect that influences not only the part of the system around where the shock occurred, but potentially the entire network. By moving away from this classical paradigm, towards a decentralized architecture with local control units, researchers have designed more flexible and resilient network systems that are less prone to systemic risks and large-scale disruptions.