A Decentralized Optimization Algorithm for Multi-Agent System Based Watershed Management

A watershed can be considered a multi-agent system (MAS) composed of water users (agents) spatially distributed in a distinct manner within a common defined environment from which they can all access a particular resource. The watershed system is characterized by distributed decision processes at the agent level and a coordination mechanism organizing the interactions among individual decision processes. This paper presents a decentralized optimization method known as constraint-based reasoning, which allows individual agents in a multi-agent system to optimize their behaviors. The method incorporates the optimization of all agents' objectives through a bargaining scheme, in which the i^th agent optimizes its objective with a selected priority for collaboration and sends the solution back to all agents that interact with it. This methodology allows agents to determine how important their own objectives are in comparison to the constraints, using a local interest factor (β_i). A large β_i value indicates a selfish agent who puts high priority on its own local benefit and ignores constraints. This analysis is applied to a hypothetical watershed with three offstream human agents (city and farms), one instream human agent (reservoir) and two ecological agents (fish habitat in the river). The proposed approach takes the objectives of individual agents into account and balances them through interactions among the agents. 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.