Study on Parallel Hybrid Evolutionary Algorithm and Its Application in RBF Networks

摘要

Based on a distributed model, multiple threads simulated, this paper put forward a new Parallel Hybrid Evolutionary Algorithm (PHEA). The method combined Simulated Annealing algorithms (SA) and some local optimizing algorithms with Genetic Algorithms (GAs). It can optimize the topology and weights of the neural networks at the same time. We propose that in parallel algorithms, the function of host and subpopulations determines corresponding method. Experience results indicate that subpopulations should enlarge search space and adopt GASA method; host should optimize individuals and adopt local optimizing method. PHEA can evolve RBFNNs efficiently. The method is more robust than non-parallel algorithms and other parallel algorithms. It is expected to apply to Data Mining and other areas.