An Exploration and Optimization of Cooperative Coevolution

Abstract

This research begins with an investigation of cooperative coevolution through comparative studies between conventional evolutionary algorithms and cooperative coevolutionary algorithms on the design of artificial neural networks. Several evaluation models developed for the cooperative coevolutionary algorithms are then analysed by applying these models to a new neuro-evolutionary algorithm developed in this research. The advantages and disadvantages between evolutionary algorithms and coevolutionary algorithms, as well as between different evaluation models are presented.

During the above work, my attention has been drawn to the limitation of current evaluation models designed for cooperative coevolution, especially for solving nonseparable problems. Although related works have made an effort to analyse and overcome this issue, no general evaluation model has been designed for cooperative coevolutionary algorithms. The next part of this research pays attention to the optimization of cooperative coevolutionary algorithms, in particular to the design of the evaluation model. A novel collaboration mechanism combined with a novel evaluation model is designed. Finally, analysis work demonstrates that the new algorithm does overcome the issue and improves the performance of cooperative coevolution for solving problems with different separabilities. The new collaboration mechanism can be generalized to any cooperative coevolutionary models