基于AQPSO的RBF神经网络自组织学习

AQPSO-based self-organization learning of RBF neural network

针对径向基函数(RBF)神经网络的结构设计及参数优化问题,提出一种自适应量子粒子群优化(AQPSO)算法.将RBF神经网络的网络规模及参数映射到粒子的空间位置,定义权值平均最优位置,从而对量子粒子群优化(QPSO)中L_(i,j)(t)进行评价,设计随粒子进化而自动调整的收缩-扩张系数β及网络规模调整规则,实现RBF神经网络结构和参数的自组织学习,提高网络的学习能力.通过非线性系统辨识以及短时交通流量预测验证所提出方法的有效性.实验研究表明,基于AQPSO的RBF神经网络自组织学习算法不仅能够获得较好的学习性能,而且其网络结构也较为紧凑.

Aiming at the structural design and parameter optimization problems of radial basis function（RBF） neural network, an adaptive quantum-behaved particle swarm optimization（AQPSO） algorithm is proposed. In order to realize the self-organization learning of RBF nerual network and improve the performance, the network size and parameters of RBF neural network are mapped to the spatial position of the particles firstly, and then the weight mean of best particle positions is defined to evaluate L_（i,j）（t） in quantum-behaved particle swarm optimization, the contraction-expansion coefficient β can be adjusted according to the evolution of particles, and the self-organization rule of neural network size Ki is introduced. Finally, as a case study, the proposed approach is applied to the nonlinear system identification and short-term traffic flow forecasting problems. Compared with other methods, it is shown that the proposed approach exhibits better results with higher accuracy and smaller size of architecture.