【摘要】 提出了RBF LBF串联网络结构、核函数个数、位置与宽度优化算法。提出了网络中核函数的分裂合并规则,在学习过程中将样本的类别信息作为指导信息,根据样本分布情况自动确定核函数个数、中心和宽度。实验表明,本方法具有分类精度高、不易陷入局部最小点的优点。更多还原

【Abstract】 An algorithm of optimally detemining the structures,number,positions and widths of kernel functions of cascade RBF-LBF networks was presented. According to the class labels of samples as supervision information,the number,center and width of kernel functions should be decided adaptively through the proposed split-and-merge operations.The cascade RBF-LBF networks as well as this learning algorithm have high classidication accuracy and good capacity to reach global minimum points,etc.更多还原