基于改进BP神经网络的可拓分类器构建  

Extension classifier construction based on improved BP neural network

作　　者：赵燕伟[1] 任设东[1] 陈尉刚[1] 楼炯炯 冷龙龙 

机构地区：[1]浙江工业大学特种设备制造与先进加工技术教育部/浙江省重点实验室,浙江杭州310014

出　　处：《计算机集成制造系统》2015年第10期2807-2815,共9页Computer Integrated Manufacturing Systems

基　　金：国家自然科学基金资助项目(51275477);国家"十二五"科技支撑计划资助项目(2012BAD10B01)~~

摘　　要：针对运用BP神经网络对可拓集进行分类时收敛速度慢且准确率低的问题,构建了一种基于改进BP神经网络的可拓分类器。由于负域和正域拥有公共边界,处于边界周围的数据会因为BP神经网络自身存在的误差而造成分类错误,为此以关联函数为基础,构建一个样本预处理函数对训练样本进行处理,使训练完的BP神经网络最后的输出结果远离公共边界;重新定义神经网络中的误差计算方法,使其符合可拓分类准则,降低输出值与期望值之间的要求以加快其收敛速度。通过螺杆空压机实例验证了该方法的有效性。To solve the problem of low accuracy and convergence rate when extensible set was classified by BP neural network, an extension classifier based on improved BP neural network was constructed. Owing to the common boundary of negative domain and positive domain, the data classification near to the boundary might be wrong caused by error of BP neural network. A pretreatment function for dealing with the data of samples based on correlation function was constructed, which made the output result of after training BP neural network keep away the bounda- ry. The error calculation of BP neural network was redefined to accord with extension classifier guidelines, which could reduce the accuracy between output and excepted value to accelerate convergence rate. Screw air compressor was taken as an example to verify the feasibility of proposed method.

关 键 词：BP神经网络 可拓分类 预处理 误差 螺杆空压机 

分 类 号：TH122[机械工程—机械设计及理论] TP391[自动化与计算机技术—计算机应用技术]