OptAgent: A Generalized Framework for the Optimization and Analysis of Agent-based Models

The Phase II project will develop prototype software OptAgent™, built on the existing foundation of OptQuest®, a commercial metaheuristic-based optimization engine, to provide a generalized framework for the analysis and optimization of agent-based models by combining optimization algorithms, agent-based simulation, and distributed processing. The software will overcome those weaknesses typically found in the use of agent based modeling technology including the difficulty in solving the model for optimality. The OptAgent software will function in three stages. In the first stage a diverse set of solutions will be created and a Markov blanket-based feature selection methodology will identify the most influential variables. In the second stage the agent-based model optimization search process will be performed. This process will be enhanced by classification and discrimination rules developed from the initial solution set using satisfiability data mining. In the final stage the best combination of agent and environment parameters will be presented along with statistics providing insights into the behavior of the model across the myriad possible combinations of inputs.