Abstract: The Hierarchical Genetic Particle Swarm Optimization Algorithm（HGAPSO） is introduced into neural network design，and then the structure and weights of neural network are optimized simultaneously.The advantage of GA to solve the discrete problem and that of PSO to solve the continuous problem are combined.Then it makes use of the steepest descent error of BP to learn the network’s weights，so global optimization and fast searching are reached.The results of “the chaotic time series prediction” demonstrate that Hierarchical Genetic Particle Swarm Optimization algorithm improves the learning performance and generalization ability of neural network to a large degree.

Key words: hierarchical genetic algorithm, particle swarm optimization algorithm, Error Back Propagation（BP） algorithm, artificial neural network, optimization, chaotic time series

摘要： 将递阶遗传粒子群算法（HGAPSO）应用于神经网络设计，可以在对网络拓扑结构优化的同时对连接权重进行求解。该算法结合了遗传算法在解决离散问题和粒子群算法在解决连续问题上的优势，并利用BP算法沿误差最速下降的能力对连接权重进一步学习，达到全局最优和快速搜索的有机结合。通过对混沌时序信号的预测，表明递阶遗传粒子群算法在较大程度上提高了神经网络的学习性能和泛化能力。

关键词: 递阶遗传算法, 粒子群算法, 误差反向传播（BP）算法, 人工神经网络, 优化, 混沌时间序列