基于FNN模型的决策算法研究  

Research on Decision Algorithm Based on Fuzzy Neural Network

作　　者：侯庆山 邢进生[1] HOU Qing-shan;XING Jin-sheng(School of Mathematics and Computer Science,Shanxi Normal University,Linfen 041000,China)

机构地区：[1]山西师范大学数学与计算机科学学院,山西临汾041000

出　　处：《计算机技术与发展》2020年第12期92-98,共7页Computer Technology and Development

基　　金：山西省软科学基金资助项目(2011041033-03)。

摘　　要：鉴于证据理论对样本分类和决策过程的复杂性以及不稳定性,提出了一种基于神经网络模型和模糊集理论的样本决策算法。为了降低样本分类和决策过程的复杂性,增强算法的稳定性和适用性,在该算法中,设计并提出了一种新的隶属度函数。应用提出的隶属函数对相关数据集样本进行模糊化处理,得到数据集的模糊化矩阵,其中输入样本数据与不同样本类别的隶属度相关联。根据隶属度矩阵,并结合性能较好的激活函数Swish-B,通过神经网络分类器,样本将被归属于特定的类。基于鸢尾花数据集对其进行可视化分析,将该方法与传统的证据理论及相关改进算法进行比较,验证了所设计的隶属度函数具有良好性能,同时实验结果证明了该算法的合理性与有效性,算法过程更为简单,鸢尾花数据集的分类准确率高达98%。In view of the complexity and instability of evidence theory in sample classification and decision-making,a sample decision algorithm based on neural network model and fuzzy set theory is proposed.In order to reduce the complexity of sample classification and decision-making process and enhance the stability and applicability of the algorithm,a new membership function is designed and proposed to blur the relevant data set samples to obtain the fuzzy matrix of the data set.The input sample data is associated with the membership of different sample categories.According to the membership matrix,the samples will be assigned to a specific class by the neural network classifier.Based on iris data set and visual analysis,this method is compared with traditional evidence theory and related improved algorithms.It is proved that the new membership function has excellent performance.Experiment shows that the proposed algorithm is reasonable and effective with simpler process.The classification accuracy of the iris data set is as high as 98%.

关 键 词：复杂性 稳定性 证据理论 样本分类 神经网络 模糊集理论 隶属度 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]