摘要
为了解决复杂产品关键质量特性(CTQs)识别问题,提出基于过滤(Filter)算法与包裹(Wrapper)算法的改进混合特征选择算法。首先应用Filter算法对复杂产品质量特性进行排序,接着应用Wrapper算法识别关键质量特性。提出一种新的方法FNO,确定Wrapper阶段所选关键质量特性数。应用过滤算法ReliefF构建混合算法ReliefF-W。算例分析表明,ReliefF-W能够有效进行CTQ识别。与一种传统混合算法相比,ReliefF-W能够在保证学习算法有较好预测精度的同时,识别出更少关键质量特性。
To identify critical-to-quality characteristics (CTQs),a hybrid algorithm of Filter and Wrapper is proposed.Quality characteristics are ranked by Filter algorithm firstly,and then Wrapper algorithm is applied to select the CTQs.A new method named FNO is proposed to calculate the best number of CTQs during the Wrapper phase.A Filter algorithm,ReliefF,is introduced to structure the hybrid algorithm named ReliefF-W.Experimental result illustrates that ReliefF-W is efficient in CTQ identification.Compared with a traditional hybrid algorithm, ReliefF-W can select fewer CTQs while ensuring high prediction accuracy.
出处
《工业工程与管理》
CSSCI
北大核心
2014年第3期53-59,共7页
Industrial Engineering and Management
基金
国家杰出青年科学基金资助项目(71225006)
国家自然科学基金资助项目(70931004)
关键词
关键质量特性
复杂产品
特征选择
过滤算法
包裹算法
RELIEFF
critical-to-quality characteristics (CTQs)
complex products
feature selectionfilter algorithm
wrapper algorithm
ReliefF