提出了一种基于径向基函数神经网络 RBFNN (Radial Basis Function Neural Networks)、模糊 C-均值聚类(FCM)算法和递归正交最小二乘法 (ROL S)的混凝土安全性专家系统评估预测方法。此方法先用 FCM算法初选多个RBFNN的函数中心 ,再采用 ROL S训练网络 ,最后结合后向选择法 ,减少初选的中心数目 ,以得到最终的有效中心值。该方法加快了 RBFNN的训练速度 ,提高了网络的运算效率。将其运用到混凝土安全性评估专家系统中 ,获得了满意的结果。将这种新算法得出的评估数据与传统的 BP网络计算出的数据进行了比较 ,进一步证明了 RBFNN及其学习算法的优越性和实用性

Evaluation Method of Concrete Security Based on RBF-FCM-ROLS ZHONG Luo,JIANG Qiong,YUAN Jing-ling,TONG Qi-wei,ZHANG Kai-song (School of Computer Science and Technology, Wuhan University of Technology, Wuhan 430070,China)The method based on RBF-FCM-ROLS was proposed. Firstly, the FCM algorithm is used to select the centers of RBFNN. Secondly the ROLS algorithm is used to train RBFNN. At last, the effective centers of RBFNN can be obtained by adopting backward selection algorithm. The satisfactory results are achieved by using this algorithm in the expert system of concrete safety evaluation and it illustrated that this method is effective and adaptive. Finally, the evaluation data, which are calculated by training RBF and BP respectively are showed and it proved that RBF and its learning algorithm, are better than that of BP.RBF neural network; expert system evaluation; ROLS algorithm; the fussy C-mean clustering