基于多RBF神经网络的板形数据建模  

Plate Data Modeling Based on Multiple RBF Neural Networks

作　　者：张秀玲[1,2] 代景欢 康学楠 李金祥 魏楷伦 ZHANG Xiu⁃ling;DAI Jing⁃huan;KANG Xue⁃nan;LI Jin⁃xiang;WEI Kai⁃lun(Key Laboratory of Industrial Computer Control Engineering of Hebei Province,Yanshan University,Qinhuangdao 066004,Hebei,China;National Engineering Research Center for Equipment and Technology of Cold Strip Rolling,Yanshan University,Qinhuangdao 066004,Hebei,China)

机构地区：[1]燕山大学河北省工业计算机控制工程河北省重点实验室,河北秦皇岛066004 [2]燕山大学国家冷轧板带装备及工艺工程技术研究中心,河北秦皇岛066004

出　　处：《矿冶工程》2019年第6期124-128,共5页Mining and Metallurgical Engineering

基　　金：河北省自然科学基金-钢铁联合研究基金项目(E2015203354);河北省高校创新团队领军人才培育计划项目(LJRC013);河北省教育厅科学研究计划河北省高等学校自然科学研究重点项目(ZD2016100);2016年燕山大学基础研究专项培育课题(16LGY015)

摘　　要：常规单RBF神经网络板形识别模型不能全面分离出输入变化对每个特征参数的影响,为此设计了多RBF神经网络板形识别模型,用多个子网络分别识别不同的特征参数,能够更直接、更充分的提取出输入与每个输出的关系。仿真研究结果表明:所设计的多RBF神经网络板形识别模型能够正确识别出全部板形缺陷的类型,并且识别精度上比单RBF神经网络板形识别模型提高了16.1%。Conventional plate recognition model with single RBF neural network cannot identify the influence of input variation on each characteristic parameter. Thus, a plate recognition model with multiple RBF neural networks was designed to solve this problem, which can identify different parameters with multiple sub-networks, and extract the relationship between each input and output more directly and adequately. The simulation results show that this plate recognition model with multiple RBF neural networks can identify all types of plate defects, with the recognition accuracy 16.1% higher compared with the plate recognition model with single RBF neural network.

关 键 词：板形识别 主成分分析 多RBF神经网络 遗传算法 

分 类 号：TG333[金属学及工艺—金属压力加工]