摘要
提出一种基于隐层优化算法的RBF神经网络预测模型——HLOA-IRBFM。在传统的免疫径向基神经网络模型(IRBFNM)的基础上引入粗糙集,将初始隐层空间进行划分。定义隐层区域密度和相对近似度等概念,提出边界区域中冗余点和孤立点的约减算法。优化后的隐层空间分布均匀,能以较少的中心数覆盖整个样本空间,弥补了IRBFNM模型过分依赖参数选取的不足。实验结果证明,HLOA-IRBFM模型比IRBFNM模型在预测性能方面具有更好的稳定性和准确性。
This paper proposes a kind of RBF neural network prediction model based on hidden layer optimization algorithm, named HLOA-IRBFM. By introducing rough set into the traditional Immune RBF Neural Network Model（IRBFNM）, the initia/ hidden layer can be classified. It offers a reduction algorithm about the redundant and isolated points by defining the hidden layer area density and relative approximation. The new hidden layer space distributes evenly and the sample space can be covered entirely with few hidden nodes, which bridges a gap of over dependence on the parameters selection of IRBFNM. Experimental result proves that prediction performance of HLOA-IRBFM is more stable and accurate than that of IRBFNM.
出处
《计算机工程》
CAS
CSCD
北大核心
2010年第18期191-193,196,共4页
Computer Engineering
基金
陕西省重点学科建设专项基金资助项目
关键词
粗糙集
RBF
神经网络
隐层优化
免疫算法
rough set
RBF neural network
hidden layer optimization
immunity algorithm
