On Structure Design for RBF Neural Network Based on Information Strength
-
摘要: 在系统研究前馈神经网络的基础上,针对径向基函数(Radial basis function, RBF) 网络的结构设计问题,提出一种弹性RBF神经网络结构优化设计方法. 利用隐含层神经元的输出信息(Output-information, OI)以及隐含层神经元与输出层神经元间的交互信息(Multi-information, MI)分析网络的连接强度, 以此判断增加或删除RBF神经网络隐含层神经元, 同时调整神经网络的拓扑结构,有效地解决了RBF神经网络结构设计问题; 利用梯度下降的参数修正算法保证了最终RBF网络的精度, 实现了神经网络的结构和参数自校正. 通过对典型非线性函数的逼近与污水处理过程关键水质参数建模, 结果证明了该弹性RBF具有良好的动态特征响应能力和逼近能力, 尤其是在训练速度、泛化能力、最终网络结构等方面较之最小资源神经网络(Minimal resource allocation net works, MRAN)、增长修剪RBF 神经网络(Generalized growing and pruning RBF, GGAP-RBF)和自组织RBF神经网络(Self-organizing RBF, SORBF)有较大的提高.Abstract: Based on the systemic investigation on the feedforword neural network, for the problem of the structure design of the RBF neural network, a new flexible structure design method is used for RBF neural network in this paper. By computing the output-information (OI) of the hidden neurons and the multi-information (MI) of the hidden nodes and output nodes, the hidden nodes in the RBF neural network can be inserted or pruned, thus the topology of the network can be modulated. This method can effectively solve the structure design of the RBF neural network. The grad-descent method for the parameter adjusting ensures the exactitude of the flexible RBF neural network (F-RBF). The structure of the RBF neural network is self-organizing, and the parameters are self-adaptive. In the end, the proposed F-RBF is used for approximating the classical non-linear functions and modelling key parameters of the wastewater treatment process. The results show that the F-RBF obtains a favorable dynamic character response and the approximating ability. Especially, comparied with the minimal resource allocation networks (MRAN), the generalized growing and pruning RBF (GGAP-RBF) and the self-organizing RBF (SORBF), the proposed algorithm is more effective in terms of training time, generalization, and neural network structure.
-
[1] Luengo J, Garcia S, Herrera F. A study on the use of imputation methods for experimentation with radial basis function network classifiers handling missing attribute values: the good synergy between RBFNs and EventCovering method. Neural Networks, 2010, 23(3): 406-418[2] Ferrari S, Bellocchio F, Piuri V, Borghese N A. A hierarchical RBF online learning algorithm for real-time 3-D scanner. IEEE Transactions on Neural Networks, 2010, 21(2): 275-285[3] Ye Jian, Ge Lin-Dong, Wu Yue-Xian. An application of improved RBF neural network in modulation recognition. Acta Automatica Sinica, 2007, 33(6): 652-654(叶健, 葛临东, 吴月娴. 一种优化的RBF神经网络在调制识别中的应用. 自动化学报, 2007, 33(6): 652-654)[4] Peng H, Wu J, Inoussa G, Deng Q L, Nakano K. Nonlinear system modeling and predictive control using the RBF nets-based quasi-linear ARX model. Control Engineering Practice, 2009, 17(1): 59-66[5] Bortman M, Aladjem M. A growing and pruning method for radial basis function networks. IEEE Transactions on Neural Networks, 2009, 20(6): 1039-1045[6] Platt J. A resource-allocating network for function interpolation. Neural Computation, 1991, 3(2): 213-225[7] Lu Y W, Sundararajan N, Saratchandran P. A sequential learning scheme for function approximation using minimal radial basis function neural networks. Neural Computation, 1997, 9(2): 461-478[8] Panchapakesan C, Palaniswami M, Ralph D, Manzie C. Effects of moving the center's in an RBF network. IEEE Transactions on Neural Networks, 2002, 13(6): 1299-1307[9] Gonzalez J, Rojas I, Ortega J, Pomares H, Fernández F J, Diaz A F. Multiobjective evolutionary optimization of the size, shape, and position parameters of radial basis function networks for function approximation. IEEE Transactions on Neural Networks, 2003, 14(6): 1478-1495[10] Feng H M. Self-generation RBFNs using evolutional PSO learning. Neurocomputing, 2006, 70(1-3): 241-251[11] Huang G B, Saratchandran P, Sundararajan N. A generalized growing and pruning RBF (GGAP-RBF) neural network for function approximation. IEEE Transactions on Neural Networks, 2005, 16(1): 57-57[12] Lian J M, Lee Y G, Scott D S, Stanislaw H Z. Self-organizing radial basis function network for real-time approximation of continuous-time dynamical systems. IEEE Transactions on Neural Networks, 2008, 19(3): 460-474[13] Qiao Jun-Fei, Han Hong-Gui. Optimal structure design for RBFNN structure. Acta Automatica Sinica, 2010, 36(6): 865-872 (乔俊飞, 韩红桂. RBF神经网络的结构动态优化设计. 自动化学报, 2010, 36(6): 865-872)[14] Barandiaran X, Moreno A. On the nature of neural information: a critique of the received view 50 years later. Neurocomputing, 2008, 71(4-6): 681-692[15] Kotaleski J H, Blackwell K T. Modelling the molecular mechanisms of synaptic plasticity using systems biology approaches. Nature Reviews Neuroscience, 2010, 11(4): 239-251[16] Han H G, Chen Q L, Qiao J F. An efficient self-organizing RBF neural network for water quality prediction. Neural Networks, 2011, 24(7): 717-725[17] Buzzi C, Grippo L, Sciandrone M. Convergent decomposition techniques for training RBF neural networks. Neural Computation, 2001, 13(8): 1891-1920[18] Neves G, Cooke S F, Bliss T V P. Synaptic plasticity, memory and the hippocampus: a neural network approach to causality. Nature Reviews Neuroscience, 2008, 9(1): 65-75[19] Xu J H, Ho D W C. A new training and pruning algorithm based on node dependence and Jacobian rank deficiency. Neurocomputing, 2006, 70(1-3): 544-558[20] Lee C Y, Lin C J, Chen H J. A self-constructing fuzzy CMAC model and its applications. Information Sciences, 2007, 177(1): 264-280
点击查看大图
计量
- 文章访问数: 2098
- HTML全文浏览量: 57
- PDF下载量: 1452
- 被引次数: 0