论文部分内容阅读
针对LM算法不能在线训练RBF网络以及RBF网络结构设计算法中存在的问题,提出一种基于LM算法的在线自适应RBF网络结构优化算法.该算法引入滑动窗口和在线优化网络结构的思想,滑动窗口的引入既使得LM算法能够在线训练RBF网络,又使得网络对学习参数的变化具有更好的鲁棒性,并且易于收敛.在线优化网络结构使得网络在学习过程中能够根据训练样本的训练误差和隐节点的相关信息,在线自适应调整网络结构,跟踪非线性时变系统的变化,使网络维持最为紧凑的结构,以保证网络的泛化性能.最后通过仿真实验验证了所提出算法的性能.
In view of the problems that LM algorithm can not train RBF network online and RBF network structural design algorithm, an online adaptive RBF network structure optimization algorithm based on LM algorithm is proposed. The algorithm introduces the sliding window and online optimization network structure, sliding window The introduction of LM algorithm not only makes the LM algorithm train the RBF network online, but also makes the network more robust to the change of learning parameters and is easy to converge.Optimization of the network structure on-line makes the network in the learning process according to the training error and Hidden nodes, adjust the network structure adaptively online and track the change of nonlinear time-varying system to maintain the most compact structure of the network to ensure the generalization performance of the network.Finally, the performance of the proposed algorithm is verified through simulation experiments.