引用本文:代伟,吴尚,南静,刘鑫.多尺度核随机配置网络的多目标回归算法[J].控制理论与应用,2026,43(4):915~926.[点击复制]
DAI Wei,WU Shang,NAN Jing,LIU Xin.Multi-target regression by multi-scale kernel stochastic configuration network[J].Control Theory & Applications,2026,43(4):915~926.[点击复制]
多尺度核随机配置网络的多目标回归算法
Multi-target regression by multi-scale kernel stochastic configuration network
摘要点击 119  全文点击 22  投稿时间：2024-01-17  修订日期：2026-02-03
查看全文  查看/发表评论  下载PDF阅读器   HTML
DOI编号  10.7641/CTA.2024.40038
  2026,43(4):915-926
中文关键词  随机配置网络  多目标回归  多核学习  多尺度核
英文关键词  stochastic configuration network  multi-target regression  multi-kernel learning  multi-scale kernel
基金项目  国家重点研发项目(2022YFB3304700), 江苏省杰出青年基金项目(BK20240102), 国家自然科学基金项目(62373361)资助.
作者单位E-mail
代伟 中国矿业大学信息与控制工程学院 weidai@cumt.edu.cn 
吴尚 中国矿业大学信息与控制工程学院  
南静 中国矿业大学信息与控制工程学院  
刘鑫* 中国矿业大学 信息与控制工程学院 15B904027@hit.edu.cn 
中文摘要
      多目标回归通过将多个回归任务的相关性纳入建模中以提高模型表现, 但是当前方法对于多目标内在更 加普遍且复杂的非线性结构关系建模效果并不理想. 本文提出多尺度核随机配置网络(MSK-SCN)构建高维输入与 多目标之间的映射以及多目标内在的非线性关系. 通过设计隔离建模机制, 使得每个子模型根据其负责的目标误差 状态完成隐含层参数配置, 进而避免多个回归任务共享模型参数导致建模质量下降; 建立多尺度核空间, 利用不同 尺度参数的核函数将低维空间数据映射到高维特征空间以增强数据的表达能力, 进而提高模型挖掘多目标间非线 性关系的能力. 基于虚拟数据集和真实数据集进行实验验证, 结果表明MSK-SCN的表现优于当前的多目标算法, 证 实了其在多目标回归问题上的有效性.
英文摘要
      Multi-objective regression improves model performance by incorporating the correlation of multiple regression tasks into modelling, but the current method is not ideal for more general and complex nonlinear structural relationships inherent in multi-target modeling. In this paper, a multi-scale kernel stochastic configuration network (MSK-SCN) is proposed to construct the correlations between high-dimensional inputs and multi-targets as well as the intrinsic nonlinear relationships of multi-targets. The isolation modeling mechanism is designed to make each submodel complete the hidden layer parameter configuration according to its responsible target error state, so as to avoid the modeling quality degradation caused by multiple regression tasks sharing model parameters. The multi-scale kernel space is established, and the kernel function of different scale parameters is used to map the low-dimensional data to the high-dimensional feature space to enhance the expression ability of the data, improving the ability of the model to mine the nonlinear relationship between multiple targets. The simulation results show that MSK-SCN performs better than the current multi-target algorithm through experiments on virtual data sets and real data sets, which proves its effectiveness in multi-target regression problems.