一种基于划分聚类和模糊神经网络的机器学习方法  

Method of Machine Learning Based on Partitioned Clustering And Fuzzy Neural Network

作　　者：谷建光[1] 张为华 王中伟[1] 解红雨[1] 

机构地区：[1]国防科技大学航天与材料工程学院,湖南长沙410073

出　　处：《系统仿真学报》2007年第23期5581-5586,共6页Journal of System Simulation

摘　　要：将基于划分的模糊聚类算法和一般模糊极小极大神经网络分类算法相结合,提出了一种新的机器学习方法,实现了基于类比的案例推理学习模型。具体实现思想是,首先利用基于确定性退火技术的划分聚类算法对已知案例进行聚类标识,由所得结果建立一般模糊极小极大神经网络分类模型,然后用该模型实现新目标问题的案例相似性检索,最后针对目标问题结果案例完成案例学习。通过实例表明,该算法具有较好性能,并在基于案例推理的固体火箭发动机总体设计中成功应用,得到了论域覆盖面大的设计结果集。Fuzzy partitioned clustering algorithm was combined with General fuzzy min-max （ GFMM） neural network, and the new machine learning was proposed. Learning by Analogy was achieved in Case Based Reasoning. The idea was a kind of partitioned clustering algorithm was designed to label old case firstly. The second, GFMM neural network for classification and case retrieving was put forward, and it was used to case retrieving based on similarity of new problem. Finally, the result case of problem was retained by case learning. Through example analysis, it＇s indicated that the new technique has good performance, and it is used in solid rocket motor system design, and the degree of design results Universe of Discourse are improved.

关 键 词：划分聚类 一般模糊极小极大神经网络 机器学习 案例推理 固体火箭发动机总体设计 

分 类 号：TP182[自动化与计算机技术—控制理论与控制工程] V435[自动化与计算机技术—控制科学与工程]