Hybrid fuzzy polynomial neural networks with the aid of weighted fuzzy clustering method and fuzzy polynomial neurons
详细信息 查看全文
- 作者：Wei Huang ; Sung-Kwun Oh ; Witold Pedrycz
- 关键词：Parallel fuzzy polynomial neural networks (PFPNNs) ; Fuzzy polynomial neural networks (FPNNs) ; Weighted fuzzy clustering method (WFCM) ; Particle swarm optimization (PSO)
- 刊名：Applied Intelligence
- 出版年：2017
- 出版时间：March 2017
- 年：2017
- 卷：46
- 期：2
- 页码：487-508
- 全文大小：
- 刊物类别：Computer Science
- 刊物主题：Artificial Intelligence (incl. Robotics); Mechanical Engineering; Manufacturing, Machines, Tools;
- 出版者：Springer US
- ISSN：1573-7497
- 卷排序：46
文摘
It is well-known that any nonlinear complex system can be modeled by using a collection of “if …then” fuzzy rules. In spite of a number of successful models reported in the literature, there are still two open issues: (1) one is not able to reflect the heterogeneous partition of the input space; (2) it becomes very difficult to deal effectively with high dimensionality of the problem (data). In this study, we present a parallel fuzzy polynomial neural networks (PFPNNs) with the aid of heterogeneous partition of the input space. Like fuzzy rules encountered in fuzzy models, the PFPNNs comprises a collection of premise and consequent parts. In the design of the premise part of the rule a weighted fuzzy clustering method is used not only to realize a nonuniform partition of the input space but to overcome a possible curse dimensionality. While in the design of consequent part, fuzzy polynomial neural networks are exploited to construct optimal local models (high order polynomials) that describe the relationship between the input variables and output variable within some local region of the input space. Two types of information granulation-based fuzzy polynomial neurons are developed for FPNNs. Particle swarm optimization (PSO) is employed to adjust the design parameters of parallel fuzzy polynomial neural networks. To evaluate the performance of PFPNNs a series of experiments based on several benchmarks are included. A comparative analysis demonstrates that the proposed model comes with higher accuracy and generalization capabilities in comparison with some previous models reported in the literature.