一种改进的RAN学习算法.PDFVIP

第 19 卷摇 第2 期摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 模式识别与人工智能摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 Vol.19摇 No.2 摇 2006年4月摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 PR AI摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 Apr摇 摇 2006 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 一一种改进的RAN 学习算法 李摇 彬 (山东大学威海分校 信息工程学院威海摇 264209) (山东轻工业学院数理学院济南摇 250100) 摘摇 要摇 提出一种资源分配网络(Resource Allocating Network,RAN)的新的学习算法,称为IRAN算法.该算法通 过一个包含4部分的新颖性准则来增加网络中的隐层神经元,通过误差下降速率来删除冗余神经元并采用基于 Givens蛳QR分解的递归最小二乘算法进行输出层权值的更新.通过函数逼近领域中2个 Benchmark 问题的仿真结 果表明,与RAN,RANEKF,MRAN算法相比,IRAN算法不但学习速度快,而且可以得到更为精简的网络结构. 关键词摇 RAN学习算法,径向基函数,隐层神经元,Givens蛳QR分解,删除策略 中图法分类号摇 TP183 An Improvement of the RAN Learning Algorithm LI Bin (School of Information Engineering,Shandong University at Weihai,Weihai264209) (College of Mathematical and Physical Sciences,Shandong Institute of Light Industry, Jinan250100) ABSTRACT An improvedlearning algorithmfor ResourceAllocating Network (RAN)ispresented inthispaper, which is called IRAN algorithm. It allocates hidden neurons of network by a four蛳part novelty criterion, removes redundantneuronsaccordingtotheirerrorreductionrates,andupdatesoutput蛳layerweightsby a recursiveleast蛳squaresalgorithmwith Givens蛳QRdecomposition. SimulationsontwoBenchmarkproblems in the function approximation area indicate that the IRAN algorithm can provide smaller networks and consume less training time than RAN,RANEKF and MRAN algorithms. Key Words摇 Resource Allocating Network (RAN) Learning Algorithm,Radial Basis Function,Hidden Neurons,Givens蛳QR Decomposition,Pruning Strategy 收稿日期:2