Graph Structure Optimization for Agent Control Problems Using ACO

Abstract

AbstractAnt Colony Optimization (ACO) is one of the powerful swarm intelligence algorithms capable of solving various problems. In this research, ACO is used to optimize individuals with graph structure. This structure is exactly like the approach taken by genetic network programming (GNP) for individual representation for solving agent control problems. However, in some types of environments such as stochastic environments, calculated fitness of an individual is not the same in each evaluation. Therefore, to estimate the true fitness of an individual, several times of evaluation are needed leading to increased process time of the evolution. In this research, a method is proposed to avoid slowing down the progress speed of the algorithm using ACO. This method can be well adapted on graph structures and in each iteration, it enhances the fitness of individuals using a constructive mechanism. In constructive mechanism, an individual is produced according to the experience of previous generations. In this research, the experience is the achieved fitness and it is distributed on the corresponding paths in the graph structure. This new method was used to solve an agent control problem called Pursuit-Domain while the environment is deterministic or stochastic. The experimental results showed high capabilities of this algorithm in generation of efficient strategies for agents in an agent control problem.KeywordsAnt colony optimizationAgent control problemsStochastic environmentsGenetic programmingGenetic network programming