摘要
模糊集理论适用于一些实验数据中不确定性和模糊性的建模问题,而模糊推理系统拥有模糊IF-THEN格式的结构化知识表示,但缺少适应性。神经网络本身具有对外部很强的适应性和从过去数据中学习的机制,但基于线性推理的模糊神经网络(FNN)模型作为模糊推理方法不能得到存在于参数间的最终关系,也不能影响接着发生的模糊集合。因此,我们提出了一个多级模糊神经网络(Multi-FNN),使用硬C均值聚类和进化模糊颗粒,利用处理为近似推理的一个线性推理,获得信息微粒和模糊集之间的关系。
Fuzzy sets theory has been introduced to model uncertain and/or ambiguous characteristics in any experi- mental data. Fuzzy inference system is expressed by the form 'if-then', but it lacks of fitness. While the essential ad- vantage of neural networks lies in their adaptive nature and mechanisms of learning from historical data. The draw- back of fuzzy neural networks(FNNs) model based on linear inference treated as fuzzy inference method is that even- tual relationships existing between the variables cannot be captured in this manner and reflected in the form of the en- suing fuzzy sets. To deal with shortcoming, we propose an idea of Multi-FNNs. They use a Hard C-Means (HCM) clustering algorithm and evolutionary fuzzy granulation and obtain relationship between information granulation and fuzzy sets by the linear inference method treated as approximation inference. The results demonstrated the effective- ness of the proposed model
出处
《计算机科学》
CSCD
北大核心
2005年第3期229-232,共4页
Computer Science
