摘要
针对目前多模型建模方法中存在的把样本点错误划分的问题,提出了一种局部信息修正与全局特征相结合的多模型建模方法。该方法确定多个模型的混合范围,引入局部差商信息,确定上述混合部分的类别归属,再此基础上利用基于最小二乘支持向量机的模糊C回归估计方法,得到多个回归模型。将该方法运用到两个仿真数据中,仿真结果表明,与现有方法相比,该方法建模正确率大大提高,均方根测试误差也有明显降低。
There exists a mis-match phenomenon in current multi-model approaches, which classify samples into undesirable clusters/groups. A new multi-model modeling approach based on incorporating local information amending and global features is proposed. First, a heuristic method is constructed to determine mixing zone. Second, local difference quotient on data in mixing zone is introduced to determine their membership for different classes. Finally, using the obtained membership, a fuzzy C regres- sion algorithm based on LS-SVM is applied to establish multiple models, Experiments on two synthetic data sets are conducted and the result show the proposed approach outperforms the traditional approach in terms of accuracy rating and RMSE on test data.
出处
《计算机工程与设计》
CSCD
北大核心
2013年第7期2551-2555,2574,共6页
Computer Engineering and Design
基金
国家自然科学基金项目(U1204609)
河南省基础与前沿技术研究计划基金项目(122300410111)
河南师范大学青年科学基金项目(2011QK25)
河南省教育厅科学技术研究重点基金项目(13A520534)
关键词
多模型
最小二乘支持向量机
差商
模糊隶属度
类归属
multi-model
LS-SVM
difference quotient
fuzzy membership
membership for different classes
