摘要
提出一种新的GMDH网络的选择性集成算法,通过对构造GMDH网络个体的训练样本进行惩罚性划分,产生具有差异性的GMDH网络集合,再利用遗传算法从中选择最优GMDH网络子集进行集成。实验结果与分析表明,与GMDH网络的整体集成和单个GMDH网络以及传统的BP神经网络集成相比,GMDH网络的选择性集成在性能上具有明显的优势。
A selective GMDH network ensemble algorithm was presented. With the punitive classification of the samples for training GMDH network individuals, a group of candidate GMDH networks were developed which were different from each other. Genetic algorithm was then used to evolve the best subset of the candidates to form the ensemble. Experiments show that compared to all-candidates GMDH ensemble and GMDH network individuals as well as the traditional BP neural network ensemble, selective GMDH network ensemble imoroves the performance greatly.
出处
《计算机应用》
CSCD
北大核心
2006年第11期2554-2557,共4页
journal of Computer Applications
关键词
GMDH
惩罚性划分
选择性集成
Group Method of Data Handling （GMDH）
punitive partition
selective ensemble
