A hyper-sphere multistep based hybrid genetic algorithm

This thesis proposes a hybrid genetic algorithm based on the hyper-sphere with multi-step. First, each class samples of population sample space is settled according to genetic characteristics of Hyper-sphere in order to separate the different classes of samples in the feature space through hyper-sphere to achieve the classification results. Then the elite population will be achieved on the classification of population computation and local optimization. Finally the elite population is put in the second genetic computing, with multi-point crossover computing using hyper-sphere retention elite seed, with mutation and evolution conducted on the different surfaces of hyper-sphere, to realize the optimization calculation of the algorithm in a very small population sample set and in a very small space cost while retaining the historical results. The simulation results show that the proposed algorithm has not only higher optimization speed, higher stability and higher accuracy compared with the traditional genetic algorithms.