堆石坝参数反演的蚁群聚类RBF网络模型.pdfVIP

第 28 卷 增 2 岩石力学与工程学报 Vol.28 Supp.2 2009 年 9 月 Chinese Journal of Rock Mechanics and Engineering Sept. ，2009 堆石坝参数反演的蚁群聚类RBF 网络模型 康 飞，李俊杰，许 青 (大连理工大学 土木水利学院，辽宁 大连 116024) 摘要：将蚁群算法与径向基(RBF) 网络相结合，提出一种用于堆石坝力学参数反演的蚁群聚类径向基网络模型。 该模型用蚁群聚类算法搜索 RBF 网络基函数中心，模拟蚁群觅食聚类的概率转移特性，所得到的聚类结果类间离 散度和比传统 K 均值聚类结果小，能够得到更合理的基函数中心，从而获得较准确的坝体参数和位移之间的非线 性映射关系。在进行参数灵敏度分析的基础，对一座堆石坝的反演分析表明，蚁群聚类 RBF 网络模型可有效地求 解堆石坝多参数反演问题，反演结果优于 BP 网络模型和K 均值 RBF 网络模型。 关键词：岩土工程；土石坝；蚁群聚类算法；RBF 网络；灵敏度；反演分析 中图分类号：TU 41 文献标识码：A 文章编号：1000 –6915(2009)增 2 –3639 –06 ANT COLONY CLUSTERING RADIAL BASIS FUNCTION NETWORK MODEL FOR INVERSE ANALYSIS OF ROCKFILL DAM KANG Fei ，LI Junjie ，XU Qing (School of Civil and Hydraulic Engineering ，Dalian University of Technology ，Dalian ，Liaoning 116024，China) Abstract ：An ant colony clustering radial basis function neural network model for parameter inverse analysis is proposed by combining the ant colony clustering algorithm with radial basis function(RBF) networks. In the new model, the radial basis function centers are searched by the ant colony clustering algorithm which utilizes the probability transfer characteristic of ant foraging clustering behavior. The sum of scatter degree obtained by the ant colony clustering algorithm is smaller than that obtained by the traditional K means clustering algorithm, thus more reasonable radial basis function centers can be searched so as to obtain the nonlinear mapping relationship between the parameters to be inversed and the displacements measured at certain points in the dam. Inverse analysis is performed to a concrete faced rockfill dam；the results show that the new neural