电子学报 ›› 2013, Vol. 41 ›› Issue (5): 897-904.DOI: 10.3969/j.issn.0372-2112.2013.05.011

• 学术论文 • 上一篇    下一篇

0阶L2型TSK迁移学习模糊系统

蒋亦樟, 邓赵红, 王士同   

  1. 江南大学数字媒体学院,江苏无锡 214122
  • 收稿日期:2012-06-04 修回日期:2012-10-22 出版日期:2013-05-25
    • 作者简介:
    • 蒋亦樟 男,1988年生于江苏无锡.2006年、2012年分别在南京理工大学、江南大学获得工学学士、硕士学位,2012年进入江南大学攻读博士学位,主要从事模式识别、人工智能、模糊系统等方面的研究. E-mail:jyz0512@163.com
    • 基金资助:
    • 国家自然科学基金 (No.60903100,No.61170122,No.61272210); 江苏省自然科学基金 (No.BK2009067,No.BK2012552)

0-Order-L2-Norm-Takagi-Sugeno-Kang Type Transfer Learning Fuzzy System

JIANG Yi-zhang, DENG Zhao-hong, WANG Shi-tong   

  1. School of Digital Media,Jiangnan University,Wuxi,Jiangsu 214122,China
  • Received:2012-06-04 Revised:2012-10-22 Online:2013-05-25 Published:2013-05-25
    • Supported by:
    • National Natural Science Foundation of China (No.60903100, No.61170122, No.61272210); Natural Science Foundation of Jiangsu Province,  China (No.BK2009067, No.BK2012552)

摘要: 针对传统模糊系统在面对源场景存在某种程度的信息缺失或所采集的信息有限导致受训所得系统泛化能力较差之挑战,探讨了具有迁移学习能力的模糊系统.该类模糊系统不仅能充分利用当前场景的数据信息,还可有效利用历史知识对当前源场景的建模过程进行辅助学习,以达到弥补信息缺失之目的.基于此思想以0阶L2型TSK模糊系统为研究对象构造了0阶L2型TSK迁移模糊系统(0-L2-TSK-TFS).在模拟数据集以及真实数据集上的实验研究亦验证了该迁移模糊系统在应对存在信息缺失的场景时,较之于传统模糊建模方法有着更好的适用性.

关键词: 迁移学习, 信息缺失, 历史知识, 0-L2-Takagi-Sugeno-Kang(0-L2-TSK)模糊系统

Abstract: When the information of partial data is missing,the classical fuzzy systems constructed based on this dataset will have the weak generalization abilities for this scene.In order to overcome this shortcoming,the fuzzy system with the transfer learning abilities,i.e.,transfer fuzzy system,is proposed.In the learning procedure,the transfer fuzzy system can learn not only from the data information in the current scene,but also from the existing useful historical knowledge.Based on this idea,a transfer learning mechanism.a specified,and L2-norm penalty based 0-order-TSK-transfer fuzzy system(0-L2-TSK-TFS)was proposed,and a transfer learning mechanism was introducde.The proposed method was verified by experiments on simulation data and real data,and shows better adaptability than traditional fuzzy modeling methods in the scene with information missing.

Key words: transfer learning, information missing, history knowledge, 0-order-L2-norm-Takagi-Sugeno-Kang fuzzy system

中图分类号: