Adaptive Parameter Selection in Evolutionary Algorithms by Reinforcement Learning with Dynamic Discretization of Parameter Range

摘要

Online parameter controllers for evolutionary algorithms adjust values ofparameters during the run of an evolutionary algorithm. Recently a newefficient parameter controller based on reinforcement learning was proposed byKarafotias et al. In this method ranges of parameters are discretized intoseveral intervals before the run. However, performing adaptive discretizationduring the run may increase efficiency of an evolutionary algorithm. Aleti etal. proposed another efficient controller with adaptive discretization. In the present paper we propose a parameter controller based on reinforcementlearning with adaptive discretization. The proposed controller is compared withthe existing parameter adjusting methods on several test problems usingdifferent configurations of an evolutionary algorithm. For the test problems,we consider four continuous functions, namely the sphere function, theRosenbrock function, the Levi function and the Rastrigin function. Results showthat the new controller outperforms the other controllers on most of theconsidered test problems.