Managing multi-agent risk and system uncertainty using options-based decision policies
The management and allocation of resources in large-scale engineering and operational systems has become increasingly complex as a result of both advances in information technology and the presence of system uncertainty. Recent technological and communication advances have provided increased access to information in large-scale systems, but have also resulted in two primary issues for decision makers. First, due to the limited capacity of resources to process this information and perform system operations, careful selection must be made to identify the information sources (and associated tasks) that should be focused on when making resource allocation decisions. Second, the vast supply of information available to an agent presents the issue of identifying when enough information has been gathered to make a decision, or if the decision should be postponed until more information has been processed. These two issues result in the need to develop an approach that identifies both where and when to allocate these finite resources. Furthermore, uncertainties about information, future conditions, and market trends may exist in various resource-constrained situations, including engineering and system design processes, new technology development, enterprise systems, homeland security, emergency preparedness and response, global operations, and supply chain management. These uncertainties provide unique challenges when developing appropriate decision-making protocols and may require the integration of risk management techniques in the decision-making process.
The primary focus of this research is to develop distributed decision policies that manage risk from multiple agent perspectives in a resource allocation system, and improve the overall performance utility for multiple agents and the system. The general approach used to govern these decisions is based on the concept of dynamic flexibility using options-based decision policies. The impact of managing system risk from a distributed decision-making perspective is evaluated with respect to improvements in both agent utilities and system properties while adhering to limited and finite capacity resource constraints.
The first part of this research considers the decision policies of agents that act in a buyer-seller manner. This model introduces the concept of using options-based policies to manage risk for a task allocation decision, tests the impact of various system parameters, and incorporates the threat of task preemption. For this initial model, the resource views tasks as providing heterogeneous profit values and, therefore, develops a pricing incentive (or disincentive) scheme that encourages (or discourages) the current task from exercising its allocation option. This pricing scheme is designed to help the resource better manage its queue strategy and have more control over its rate of revenues. Because both the task and resource agents are making decisions to maximize their individual profits, utility is transferred between these agents and the resulting policies yield a zero-sum game.
The second part of this research extends this flexible, options-based approach to hedge risks posed by the underlying system uncertainties for both the task and resource agents, and explores the endogenous relationships between agent decisions and the evolution of these uncertainties. The option of allocating a task to a resource for processing is valued from a multi-agent perspective and a risk-based, distributed decision-making policy is developed that improves the utilities of both the task and resource agents. Because the actions of decision makers may have an impact on the evolution of the underlying source of uncertainty, this relationship is modeled and a solution approach developed that converges to an equilibrium system state. The final result is a distributed decision-making policy that both responds to and controls the evolution of risk due to uncertainty in a resource allocation system. The theoretical advances obtained from this work can be extended to other problem domains within industrial engineering and operations research that would benefit from enhanced flexibility in the decision-making process.