摘要
提出一种改进的BP神经网络处理板形缺陷数据的方法,建立双隐层BP神经网络模型,并对Sigmoid激活函数的形状进行调节。将其应用到冷轧的板形缺陷识别中,与利用Levenberg-Marquardt规则训练的BP神经网络预测结果作对比,表明该方法不仅有效地减少双隐层BP网络的学习时间,同时改善了网络的泛化能力,有利于板形缺陷在线识别。
In order to build a double hidden-layer BP neural network model to adjust the shape of Sigmoid activation function,a method improving the BP neural network to preprocess the plate defective data was proposed.Comparing the data of this method with the formula Levenberg-Marquardt preprocessing method,the results show the time of learning BP neural network can be effectively reduced and the network＇s generalization ability be improved by this method,and it benefits the on-line identification of the plate defects.
出处
《化工自动化及仪表》
CAS
北大核心
2010年第4期42-44,48,共4页
Control and Instruments in Chemical Industry
基金
国家自然科学基金资助项目(60973042)
山东省自然科学基金资助项目(Y2008G20)
关键词
板形识别
双隐层BP神经网络
SIGMOID函数
L-M优化算法
flatness recognition
double-layer BP neural network
Sigmoid function
Levenberg-Marquardt optimization algorithm
