Study of policies for agent collaborative decision-making in risk assessment and management of infrastructure networks

摘要

Physical infrastructure networks are complex distributed systems which are vital for society. The human aspect of decision-making processes makes infrastructure engineering a socio-technical problem, which is commonly carried out in a decentralised way, particularly during major disruptions. The authors previously introduced the Complex Distributed Agent Network (CoDAN) framework for risk-based decision-making in complex systems. When analysing infrastructure systems with CoDAN, traditional physical and network features are combined with organisational decision-making processes involved in their operation. The latter is achieved by assigning agent capabilities to certain sub-units in the network, which are detected via pattern recognition techniques that provide reasonable physical partitions. CoDAN explored coordination amongst agents but did not consider cooperation, which may lead to improvements. This paper explores collaboration amongst CoDAN agents associated to infrastructure network sub-units, which perform risk assessment and make quasi-optimal decisions about network operation. The contribution of this paper is to allow and study information exchange policies amongst agents, determining the implications of which information agents share and with whom they share it. Illustrative examples suggest that interactions amongst agents frequently leads to improvements in their optimisation objectives (e.g., enhancing reliability); further improvements can be obtained by intelligently selecting with which agents to collaborate. For practical applications, it is essential to consider the cost for agents to transmit and process information as well as the possible obstacles to implement such policies in real environments.