Adaptive parameter tuning for agent-based modeling and simulation

摘要

The purpose of this study was to solve the parameter-tuning problem of complex systems modeled in an agent-based modeling and simulation environment. As a good set of parameters is necessary to demonstrate the target behavior in a realistic way, modeling a complex system constitutes an optimization problem that must be solved for systems with large parameter spaces. This study presents a three-step hybrid parameter-tuning approach for agent-based models and simulations. In the first step, the problem is defined; in the second step, a parameter-tuning process is performed using the following meta-heuristic algorithms: the Genetic Algorithm, the Firefly Algorithm, the Particle Swarm Optimization algorithm, and the Artificial Bee Colony algorithm. The critical parameters of the meta-heuristic algorithms used in the second step are tuned using the adaptive parameter-tuning method. Thus, new meta-heuristic algorithms are developed, namely, the Adaptive Genetic Algorithm, the Adaptive Firefly Algorithm, the Adaptive Particle Swarm Optimization algorithm, and the Adaptive Artificial Bee Colony algorithm. In the third step, the control phase, the algorithm parameters obtained via the adaptive parameter-tuning method and the parameter values of the model obtained from the meta-heuristic algorithms are manually provided to the developed tool performing the parameter-tuning process and they are tested. The best results are achieved when the meta-heuristic algorithms that were successful in the optimization process are used with their critical parameters adjusted for optimum results. The proposed approach is tested by using the Predator-Prey model, the Eight Queens model, and the Flow Zombies model, and the results are compared.