Scholarlink

<p>In this paper, an evolving hierarchical fuzzy system based on variable selection and variable expected output (EHFS-VSVO) is proposed to address the challenges of dynamic, diverse and high speed data streams. The system combines the advantages of online learning in evolving fuzzy systems and hierarchical fuzzy systems, capable of handling high-dimensional complex problems. In the single-layer fuzzy system, an improved dynamic threshold scheme for fuzzy rule generation is proposed in EHFS-VSVO. This scheme not only takes into account the model’s online training error but also dynamically regulates the rule generation rate according to the number of rules. To delete fuzzy rules more cautiously, a metric based on the frequency of rule usage is introduced in EHFS-VSVO to assess whether the rules have become obsolete, in addition to considerg the utility value of the rules. In addition, EHFS-VSVO simplifies the complexity of the model through variable selection, ordering, and a hierarchical structure with variable expected output cascades. The performance advantages of EHFS-VSVO are demonstrated through simulation examples using multiple real-world datasets, particularly in terms of accuracy and generalization ability when dealing with online regression problems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An Evolving Hierarchical Fuzzy System Based on Variable Selection and Variable Expected Outputs

  • 期刊论文

  • 发表日期：

    2025年7月20日

查看全文
  • Pengcheng Wang,
  • Tao Zhao,
  • Hainan Yang,
  • Jian Zhang

摘要

In this paper, an evolving hierarchical fuzzy system based on variable selection and variable expected output (EHFS-VSVO) is proposed to address the challenges of dynamic, diverse and high speed data streams. The system combines the advantages of online learning in evolving fuzzy systems and hierarchical fuzzy systems, capable of handling high-dimensional complex problems. In the single-layer fuzzy system, an improved dynamic threshold scheme for fuzzy rule generation is proposed in EHFS-VSVO. This scheme not only takes into account the model’s online training error but also dynamically regulates the rule generation rate according to the number of rules. To delete fuzzy rules more cautiously, a metric based on the frequency of rule usage is introduced in EHFS-VSVO to assess whether the rules have become obsolete, in addition to considerg the utility value of the rules. In addition, EHFS-VSVO simplifies the complexity of the model through variable selection, ordering, and a hierarchical structure with variable expected output cascades. The performance advantages of EHFS-VSVO are demonstrated through simulation examples using multiple real-world datasets, particularly in terms of accuracy and generalization ability when dealing with online regression problems.