A decentralized optimization algorithm for multiagent system-based watershed management

A watershed can be simulated as a multiagent system (MAS) composed of spatially distributed land and water users (agents) within a common defined environment. The watershed system is characterized by distributed decision processes at the agent level with a coordination mechanism organizing the interactions among individual decision processes at the system level. This paper presents a decentralized (distributed) optimization method known as constraint-based reasoning, which allows individual agents in an MAS to optimize their behaviors over various alternatives. The method incorporates the optimization of all agents' objectives through an interaction scheme, in which the ith agent optimizes its objective with a selected priority for collaboration and forwards the solution and consequences to all agents that interact with it. Agents are allowed to determine how important their own objectives are in comparison with the constraints, using a local interest factor (beta(i)). A large beta(i) value indicates a selfish agent who puts high priority on its own benefit and ignores collaboration requirements. This bottom-up problem-solving approach mimics real-world watershed management problems better than conventional "top-down'' optimization methods in which it is assumed that individual agents will completely comply with any recommendations that the coordinator makes. The method is applied to a steady state hypothetical watershed with three off-stream human agents, one in-stream human agent (reservoir), and two ecological agents.