.png){kind=icon}
.png){kind=icon}
Return
Evolving kernel-based fuzzy system with nonlinear consequences
.png){kind=icon}
PRE
AI
DOI：10.1016/j.asoc.2024.112384.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The fuzzy system equipped with nonlinear consequences poses significant nonlinear generalization capabilities. Few works are presented on evolving fuzzy system with nonlinear consequences. This paper presents an evolving kernel-based fuzzy system (EKFS) with nonlinear consequences considering the powerful modeling ability of nonlinear consequences and self-evolving learning ability. The major novelties of our work are: (1) employing the measurement of kernel similarity and approximation error to obtain an appropriate structure with the strategies of rule recruiting and rule shrinking during the adaptive learning phase instead of structure predefining, (2) designing the rule shrinking strategy which is indeed a dimensionality reduction technology for the kernel vector to calculate an equivalent low dimensional expression without losing approximation accuracy, (3) achieving the system structural lightweight and online performance improvement via the structure evolution throughout the approximation process. Through the real prediction problems drawn from the Time Series Data Library and Typhoon tracking problem, the proposed EKFS demonstrate better prediction performance in terms of Nonlinear Auto-Regressive with eXogenous inputs model compared with the existing state-of-art methods.
Keywords:
Evolving
Rule recruiting
Rule shrinking
Time series prediction
Journal
IF:
6.6
Papers:
1.4W
Citations:
4.8W