模糊神经网络建模方法的研究

Studies on the Modeling of Fuzzy Neural Networks

近年来神经网络在建模中得到了广泛地应用 ,但其学习过程需要大量的训练样本以保证其结果的正确性。在工业过程建模中 ,神经网络因可采集与训练样本数少 ,且信息不全等困难 ,难以建立一定正确度的模型。针对这一问题 ,本文以Gauss函数为隶属度函数形式改进模糊聚类的C -平均法 ,提出了模糊CG -平均法 ,对一同组数据的聚类结果证明了此方法的有效性。模糊神经网络在化工中的应用研究尚处于初级阶段 ,本文将模糊CG-平均法与神经网络结合 ,构造由模糊化层、隶属度生成层、推理层及反模糊化输出层构成的模糊神经网络 ,实例表明本文所构造的模糊神经网络在使用较少训练样本的条件下仍能取得理想的结果 。

There are difficulties of Neural Networks (NN) in establishing a correct model because of the lack of information and samples that can be collected in process and used in training. In order to overcome these difficulties, this article use Gauss function as membership function to improve the C-average method, a fuzzy clustering method. Furthermore it is combined with NN to establish Fuzzy Neural Networks (FNN). The examples show that the FNN of this article can get ideal results even using fewer training samples, which is helpful to establish model from production directly.