基于GA的GMDH选择性集成及在煤炭价格系统模型中的应用  

Selective ensemble of GMDH models via genetic algorithm and its application to coal price modeling

作　　者：王晶[1] 李增光[1] 

机构地区：[1]南京航空航天大学自动化学院,南京210016

出　　处：《系统工程理论与实践》2014年第5期1243-1248,共6页Systems Engineering-Theory & Practice

基　　金：国家自然科学基金(20806040);国家自然科学基金重点项目(61034005)

摘　　要：首先给出了一种通过对样本数据进行惩罚性划分,产生彼此之间具有差异性的GMDH学习器集合,然后利用遗传算法从已产生的GMDH个体集合中选择最优的个体进行选择性集成的算法,并将该方法应用于煤炭价格系统的实践研究.结果表明这种采用遗传算法选择性集成惩罚性GMDH个体的算法,与单个GMDH算法和采用遗传算法选择性集成随机性GMDH个体的算法相比,明显提高了模型的泛化能力和稳定性.该方法很好地给出了煤炭价格系统的模型,能够准确预测煤炭价格的变动趋势.An algorithm for selective ensemble of group method of data handing （CMDH） models is presented and applied to coal price modeling, where the training samples are first divided into groups with punitive action, to generate a set of diversify GMDH learner candidates, and then genetic algorithm （GA） is used to determine the optimal candidate for selective ensemble. Compared with the existing algorithms of traditional single GMDH model and the random ensemble of GMDH models, the proposed algorithm for selective ensemble of GMDH based on punitive division can significantly improve the model generalization capability and stability. Based on the proposed algorithm, a reasonable coal price model is obtained, which can provide accurate prediction of the coal price variation.

关 键 词：GMDH 遗传算法 选择性集成 惩罚性划分 煤炭价格预测 

分 类 号：N945.12[自然科学总论—系统科学]