Abstract
The problem of setting suitable parameters for population-based Evolutionary Algorithms (EA) is not new. However, the process of tuning the EA parameters is still challenging, since their sensitivity to the given problem is highly non-linear. This paper proposes a framework that uses Particle Swarm Optimisation (PSO) based on Surrogate Modelling (SM) to optimise population-based EA parameters before they can be applied to solve problems. The proposed framework is comprised of two components; PSO that searches the parameters space and a Radial Basis Function Networks (RBFN) surrogate model to guide it. The main advantage of our model is that it optimises the EA parameters in a way that ensures that EA searches the problem within a limited number of evaluations. Experiments with three different benchmark problems demonstrate that our proposed framework managed to assist a Genetic Algorithm (GA) in order to optimise its parameters and achieves better solutions than the use of Standard PSO without surrogate assistance to optimise the GA parameters, Standard GA that is applied directly to the problem with fixed parameters settings, Standard 1+1 Evolutionary Strategy (ES) applied directly to the problem and simple Random Search.