模糊数据流的进化粒度神经网络分类算法

Classification algorithm of the evolving granular neural network for fuzzy data streams

模糊数据流的分类问题大多从模糊数据流中提取典型的特征来进行分类,没有考虑到概念漂移及非平衡问题。基于此,从模糊粒度神经元入手,构建了进化粒度神经网络的多层次拓扑结构。采用了模糊神经元的信息聚集规则,提出了进化粒度神经网络的模糊编码方法与快速进化原理。运用梯形隶属函数对进化粒度神经元的聚集和模糊推理功能进行递归,通过关联函数和核函数来评估奇异逼近与粒度的近似结果,并以进化迭代和半监督分类方法解决了模糊数据流中的概念漂移及非平衡问题,从而实现了对模糊数据流的有效分类,仿真结果也证明了该方法的有效性。

Most previous research has classified fuzzy data flow on the basis of some typical features extracted from the fuzzy data flow and fails to consider the problems of concept drifting and imbalance. We used the fuzzy granularity neuron to construct a multilevel topology structure of an evolving granular neural network. Based on the information gathering rule,we propose a fuzzy encoding method and rapid evolution theory of the evolving granular neural network. In addition,we used the trapezoidal membership function to gather the evolving granular neurons and the recurring fuzzy reasoning function. We used correlation and kernel functions to evaluate the singular approximation and the approximate granularity results. We also used evolutionary iteration and the semi-supervised classification method to solve the concept drifting and imbalance problems of fuzzy data flow,in order to effectively classify the fuzzy data flow. The simulation results indicate that this method is reasonable and correct.