摘要
为提高复杂工业过程中某些关键参数的预测精度,提出一种基于改进流形正则化随机配置网络(improved manifold regularization stochastic configuration networks,IMRSCNs)的软测量建模方法。该方法首先采用基于迁移学习的特征提取思路,集成最大方差、协方差分布差异和最大均值差异获取特征变换矩阵,将训练集和测试集的特征信息投影到一个公共子空间。进一步将子空间的训练集数据输入带有流形正则化的随机配置网络中,训练网络模型,以保持数据在原特征空间的几何结构。通过多组实验结果表明,相较于原始随机配置网络(stochastic configuration networks,SCNs),所提的改进流形正则化SCNs模型拥有更高的预测精度和更好的泛化性能。
To improve the prediction accuracy of certain key parameters in complex industrial processes,a soft sensor modeling method based on improved manifold regularization stochastic configuration networks is proposed.Firstly,the feature extraction idea based on transfer learning is adopted to obtain the feature transformation matrix by integrating the maximum variance,covariance distribution difference and maximum mean difference,and then the feature information of the training set and the test set is projected into a common subspace.Then the training data in the subspace is further input to the stochastic configuration network with manifold regularization to train the network model,and to maintain the geometric structure of the data in the original feature space.It is shown by several sets of experimental results that the proposed improved SCNs model with manifold regularization has higher prediction accuracy and better generalization performance.
作者
杜知微
张凤南
杨海
DU Zhi-wei;ZHANG Feng-nan;YANG Hai(School of Mechanical Engineering and Automation,Shenyang Institute of Technology,Shenfu Reform and Innovation Demonstration Zone 113122,China)
出处
《控制工程》
CSCD
北大核心
2023年第5期872-880,共9页
Control Engineering of China
基金
辽宁省自然科学基金重点领域联合开放基金资助项目(2021-KF-11-05)。
关键词
流形正则化
随机配置网络
公共子空间
软测量建模
Manifold regularization
stochastic configuration networks
common subspace
soft sensor modeling