一种随机配置网络软测量模型的稀疏学习方法
DOI:
作者:
作者单位:

1.盐城工学院;2.中国矿业大学

作者简介:

通讯作者:

中图分类号:

TP18

基金项目:

国家自然科学基金(62003293, 61973306,62003292),江苏省自然科学基金(BK20191043,BK20200086),盐城工学院校级科研项目资助(xjr2019018),东北大学流程工业综合自动化国家重点实验室开放课题(2020-KF-21-10)资助.


A Sparse Learning Method for SCN Soft Measurement Model
Author:
Affiliation:

1.Yancheng Institute of Technology;2.China University of Mining and Technology

Fund Project:

Supported by the National Natural Science Foundation of China (62003293, 61973306,62003292), the Natural Science Foundation of Jiangsu Province(BK20191043,BK20200086), the Funding for School-Level Research Projects of Yancheng Institute of Technology (xjr2019018) and the Open Project Foundation ofState Key Laboratory of Synthetical Automation for Process Industries (2020-KF-21-10).

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    随机配置网络 (Stochastic configuration network, SCN) 构建一个不等式约束条件对隐性参数进行随机分配, 同时对其范围进行自适应选择, 具有收敛速度快、建模精度高等优点. 由于随机算法的本质特性, 不可避免产生低值、冗余节点, 为提高 SCN 软测量模型的稀疏性, 本文提出一种简约随机配置网络 (Parsimonious stochastic configuration network, PSCN). PSCN 在网络增量构建目标函数中引入 L1 范数, 建立一个新的不等式约束条件来保障隐性节点 的生成质量. 并进一步, 针对新建目标函数的非凸性和非光滑性, 采用交替方向乘子法 (Alternating direction method of multipliers, ADMM) 对整个 PSCN 网络的输出权重进行更新. 最后, 将本文方法应用于基准数据集和实际工业过程软测量问题中, 证明其可有效简化模型结构, 同时具有较高的泛化性能.

    Abstract:

    For stochastic configuration network (SCN), it randomly produces the hidden parameters and adaptively selects their scopes using an inequality constraint. As a result, SCN exhibits superior performance in convergence speed and modeling accuracy. It is inevitable to produce low-value and redundant hidden nodes due to the inherent feature of randomized algorithm. To improve the sparsity of SCN soft sensor model, a parsimonious stochastic configuration network (PSCN) is proposed in this paper. L1 norm is plugged into the cost function of PSCN, and a new inequality constraint is built to obtain the high-quality hidden nodes. Next, considering the non-smoothness and non-convexity of cost function with L1 norm, the alternating direction method of multipliers (ADMM) is employed to update the output weights of whole network. Finally, the proposed method is applied to benchmark data sets and soft measurement issue in industrial process, and simulation results show that it can effectively simplify the network structure and possess the higher generalization.

    参考文献
    相似文献
    引证文献
引用本文
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2021-06-17
  • 最后修改日期:2021-08-10
  • 录用日期:2021-08-18
  • 在线发布日期:
  • 出版日期: