基于超图正则化的域适应偏最小二乘多工况软测量模型

Multi-condition soft sensor modeling of domain adaptation partial least squares based on hypergraph regularization

中文关键词  多工况  超图  结构保持  域适应  软测量

英文关键词  multiple working conditions  hypergraph  structure preservation  domain adaptation  soft sensor

中文摘要

      针对流程工业中, 因多工况导致数据分布变化引起传统软测量模型预测性能恶化问题, 本文提出一种基于 超图正则化的域适应多工况软测量回归模型框架. 首先, 采用非线性迭代偏最小二乘回归算法为基模型, 在潜变量 空间利用历史工况数据重构当前工况数据, 以增强工况间的相关性, 有效减小数据分布差异; 同时, 对重构系数施加 低秩稀疏约束, 保留了数据的局部和全局子空间结构; 其次, 通过超图拉普拉斯正则项对域适应潜变量求解过程进 行约束, 避免在寻找潜变量过程中破坏数据结构. 最后, 利用交替方向乘子法优化求解模型参数. 在多个数据集上 的实验表明, 本文方法在多工况环境下可有效提高软测量模型的预测精度和泛化性能.

英文摘要

      Multiple conditions in industrial processes can lead to changes in data distribution, which in turn can cause traditional soft sensor models to become inaccurate. Therefore, this paper proposes a domain-adaptive multi-conditions soft sensor regression model framework based on the hypergraph regularization. First, the nonlinear iterative partial least squares algorithm is used as the basic model to reconstruct the current condition data by using historical condition data in the latent variable space, to enhance the correlation between conditions and effectively reduce the differences in data distribution; Meanwhile, a low-rank sparsity constraint is imposed on the reconstructed coefficients to preserve the local and global subspace structure of the data; Secondly, the domain-adaptive latent variable solving process is constrained by the hypergraph regularterm, which effectively avoids the data structure being destroyed in the process of searching for latent variables. Finally, the model parameters are optimized by using the alternating direction multiplier method. Experiments on multiple datasets show that the method can effectively improve the prediction accuracy and generalization performance of the soft sensor model under multiple working conditions.