Modeling And Application Of Sustainable Evolutionary Algorithm Based On GP

This paper proposed the sustainable evolutionary algorithm—HFC (Hierarchical Fair Competition) model which improves the flaw of standard genetic algorithm. Inspired by the fair competition principle in biologic systems, the HFC model is developed. In the HFC model, the individual are divided into several fitness grades by different fitness and keep evolution going at all fitness grades. And there are the admission threshold and output threshold to control the movement of individuals between two grades. HFC transforms the conventional convergent evolutionary computation model into a sustainable search framework. HFC model can not only relieve the problem of premature convergence, but also ensure the diversity of population. But because the HFC model is developed on the basis of the standard genetic algorithm, there is the inherence flaw on its structure in the HFC model. In order to support the sustainable evolution better, the structure of genetic programming (GP) is added to the HFC model, in addition, some adaptive mechanism are also used in HFC model. Thus, three kinds of improved HFC algorithm are proposed in this paper.The three kinds of HFC algorithms are static HFC algorithm (SHFC), HFC algorithm with adaptive admission threshold (HFC-ADM) and HFC algorithm with adaptive migration structure (HFC-ATP). the SHFC algorithm provides the method of floating subpopulation to advance the search ability of HFC algorithm; the HFC-ADM algorithm can adjust the admission thresholds of every grade adaptively; the HFC-ATP algorithm allows the dynamic movement of individuals between certain grades. Furthermore, for guaranteeing the right evolution of topology in GP, structure fitness sharing (SFS) is proposed based on the fitness sharing method. SFS method combines the open-ended topology structure exploration of GP and the parameter optimization way, which can implement the topology search and parameter search at the same time.Finally, this paper describes the detailed algorithm process of four kinds of HFC model—HFC, SHFC, HFC-ADM and HFC-ATP. And the paper proves the feasibility of HFC algorithm and improved HFC algorithm by the wire problem and traveling salesman problem respectively. At the same time, it is proved that these HFC algorithms based on GP realize the sustainable evolution.