Smart Infrastructure: Autonomic management of urban-scale critical infrastructure

Motivation

Large-scale critical infrastructures (LSCI) are facilities, services and installations that are essential to the functioning of a society or economy. The failure of these systems could result in loss of life, significant property damage, or damage to the environment. Examples include the electricity, water and gas supply systems, phone networks (both mobile and fixed-line), data networks, transport systems, health services, business supply chains, etc. By definition, LSCIs are large in scale, spanning considerable distances (cities, countries or even continents) and/or comprising a large number of components. These systems are characterized by ever-changing operational conditions such as fluctuating supply and demand and individual components going on- and off-line.

Traditional centralised and static approaches to the management of resources and infrastructures do not cope well with the scale, the complexity and dynamicity of cities’ critical infrastructures as they are too complicated to model a priori. For this reason, we believe that smart infrastructures must be autonomic, i.e., capable of managing themselves based only on high-level objectives given by humans. In such systems the details of how to meet their objectives, even in the face of changing operating conditions, are left to the systems themselves. Therefore, autonomic systems are required to be able to self-optimize, self-heal, self-protect, and self-configure. Enabling autonomic behaviour is particularly challenging in decentralized systems where central control is not tractable or even possible, due to the large number and geographical dispersion of the entities involved.

Scientific basis

Due to the large-scale and dynamic nature, the specific actions that ensure the optimal management of urban critical infrastructure cannot be predefined but must be learnt during system operation. Such systems are often implemented as groups of agents that self-organize based only on local actions and interactions, so that the global behaviour of the system, required to meet its objectives, emerges from the agents' local behaviours. We investigate self-organizing learning algorithms for decentralized cooperative multi-agent optimization towards multiple policies in autonomic systems, learning approaches for dealing with uncertainty of the environment representation, and learning of dynamic application-sensor relationships.