面向半监督数据集的智能软测量建模方法研究与应用

According to the semi-supervised data characteristic in the process industry, this project intends to carry out research on intelligent soft sensor modeling based on the theories of Bayesian learning and semi-supervised learning. First, the semi-supervised data characteristic of the process industry is analyzed. Second, detailed analyses of advantages/disadvantages and transformation between supervised and semi-supervised soft sensor models are provided. Third, based on the active modeling structure, several intelligent data analysis methods are extended to semi-supervised forms, including radius basis active learning algorithm, nonlinear active learning algorithm, kernel-based methods, etc. Fourth, in order to handle complex data characteristics in the process industry, the Bayesian statistical learning method is introduced for semi-supervised soft sensor modeling, including mixture Bayesian linear model, nonlinear probabilistic model, semi-supervised adaptive probabilistic model, etc. Finally, robustness, suitability, and reliability of the built semi-supervised soft sensor models are evaluated and analyzed under the ensemble learning framework. The aim of this project is to propose a set of soft sensor modeling methods for semi-supervised datasets in the process industry. The feedback quality control performance, modeling time and resource, operational efficiency, and product development cycle will all benefit from this project in both theory and application aspects.

针对流程工业过程中普遍存在的半监督数据结构特征，本项目拟开展基于贝叶斯和半监督学习理论的智能软测量建模方法研究。具体研究内容包括：分析流程工业过程半监督数据结构特点；有监督和半监督软测量模型相互转化条件和优缺点分析；在主动学习建模框架下，研究基于智能数据统计分析模型的半监督软测量方法，包括径向基线性主动学习算法、非线性主动学习算法以及核学习半监督软测量建模方法等；面向工业过程的复杂数据特性，研究基于贝叶斯统计学习的半监督概率软测量建模方法，包括混合贝叶斯线性模型、非线性概率模型以及半监督迭代概率模型等；融合集成学习算法，对半监督软测量模型的鲁棒性、适应性以及可靠性进行分析。项目的研究目标是提出一套适合流程工业半监督数据集的软测量建模方法。研究成果对提升过程产品质量的闭环控制性能、节省建模时间与成本、提高过程运行效率以及缩短产品开发周期均具有重要的理论意义与实际应用价值。

针对流程工业过程中普遍存在的半监督数据结构特征，本项目开展了基于贝叶斯和半监督学习理论的智能软测量建模方法研究。具体研究内容包括：分析流程工业过程半监督数据结构特点；在主动学习建模框架下，研究基于智能数据统计分析模型的半监督软测量方法，面向工业过程的复杂数据特性，研究基于贝叶斯统计学习的半监督概率软测量建模方法，包括混合贝叶斯线性模型、非线性概率模型以及半监督迭代概率模型等；融合集成学习算法，对半监督软测量模型的鲁棒性、适应性以及可靠性进行分析。项目提出一套适合流程工业半监督数据集的软测量建模方法。研究成果对提升过程产品质量的闭环控制性能、节省建模时间与成本、提高过程运行效率以及缩短产品开发周期均具有重要的理论意义与实际应用价值。