【作者】 周鹏； 杨铸；

【Author】 Peng ZHOU1,Zhu YANG2 1.State Key Laboratory of Optical Communication Technologies and Networks,Wuhan Research Institute of Posts & Telecommunications,Wuhan,China 2.State Key Laboratory of Optical Communication Technologies and Networks,Wuhan Research Institute of Posts & Telecommunications,Wuhan,China

【机构】 武汉邮电科学研究院,光纤通信技术和网络国家重点实验室；

【摘要】 运用自适应非正交方法构建径向基函数(RBF)神经网络模型预测混沌时间序列。在RBF网络模型构造非监督学习过程中,提出了一种能同时确定了径向基函数的中心个数和对应的扩展常数的自适应方法。隐层节点在非正交空间中递推选择,扩展常数则以贝叶斯准则(BIC)收敛值作为代价函数利用微分进化(DE)算法取得。为了验证网络的预测性能,将该方法对Logistic和Henon混沌时间序列进行了仿真研究,仿真结果表明了该方法的有效性。更多还原

【Abstract】 An automatic and on-orthogonal learning algorithm of radial basis function(RBF) neural network is considered for chaotic time series prediction.During RBF neural network construction with the unsupervised learning,a new adaptive algorithm is presented to specify the locations and widths of kernels simultaneously.Hidden units are selected recursively based on RBF energy in the non-orthogonal space,and the best kernel width is obtained by Differential Evolution(DE) method that Bayesian information criterion convergence value is used as a cost function.To test the validity of model prediction,the proposed algorithm is applied to Lorenz,Logistic and Henon chaotic time series prediction,and the simulation result show that it can do well.更多还原