An Architecture for Modelling Individual Behaviour and Landscape Scale Outcomes in an Intelligent Agent-Based Simulation of Environmental Management

An Architecture for Modelling Individual Behaviour and Landscape Scale Outcomes in an Intelligent Agent-Based Simulation of Environmental Management

This work describes a software architecture that couples intelligent agent technology with Geographic Information System (GIS). The agents simulate individual behaviour, while GIS-based simulation models represent the landscape that is observed and acted upon by the agents. Our motivation in developing the software is to assist in understanding and resolving complex, semistructured, environmental management problems, where multiple actors participate in decision-making and management. Like real stakeholders, agents are individuals and have different knowledge and interests. The agents’ design is based upon the Belief, Desire, and Intention (BDI) theory of intelligent agents and implemented in Visual Basic. Agents are initialised with a set of goals they want to achieve and a library of plans that stores knowledge about landscape management actions. Each plan is like a ‘if-then’ rule where the ‘if’ is a set of conditions that need to be matched to agents beliefs for execution and the ’then’ part, which describes an outcome when the plan is executed. Agents act by looking for plans that match their goals and their current beliefs. At each time step, for each outstanding goal they examine their plan library and choose the best plan, if any, which satisfies the goal. Plan choice is influenced by the agent’s current belief state and values. Plans can be linked together in a hierarchy providing flexible and responsive reasoning to the agents. Landscape scale outcomes of the collective agent’s actions are simulated using GIS-based simulation models. Agents can have goals and plans that relate to the system wide state simulated by the GIS models and can observe changes in system state. By comparing these changes with existing conditions agents can determine if catchment resources are degrading or improving. If there is a degradation in one or more landscape resources the agent will act to revise its management actions in an attempt to improve its performance and if the goal is a system wide resource the performance of the catchment as a whole. This process is iterated until no improvement in system performance is detected. We demonstrate the potential of this decision-making framework on an abstract landscape.