基于混合遗传算法的自适应神经网络优化设计

传统遗传算法优化神经网络存在"近亲繁殖"、"早熟收敛"、收敛速度慢和容易陷入局部极小等缺点.将适应度与相应的个体数目相联系,提出一种自适应交叉变异概率,并将其用于遗传操作,使得个体具有较强的多样性,一定程度缓解种群"早熟";将单纯形法和遗传算法结合到一起,使遗传算法的搜索更具有方向性,提高遗传算法的搜索能力,加快收敛速度.仿真实验进一步证明本文提出的算法对加快收敛速度,防止"近亲繁殖",保持种群多样性比较有效.

An Adaptive Neural Network Optimization Based on Hybrid Genetic Algorithms

There are such defects in the traditional genetic algorithms as inbreeding,prematurity,slow convergence speed and easy orientation to the local minimum.Through the improving of crossover probability and mutation probability,the diversity of the network could be maintained and it avoids prematurity to some extent.The combination of the simplex method and the genetic algorithm makes the genetic algorithm search more directional and improves the search ability of genetic algorithms.The experiments show that th...