基于动态隐变量模型的工业过程动态软测量建模及应用

Soft sensors play an important role in predicting the difficult-to-measure quality variables and parameters in industrial processes, which can provide important information for proces monitoring, control and optimization. As static soft senors cannot describe the process dynamics and output-input relationships accurately, it is necessary to carry out dynamic soft sensor modeling. However, the prediction accuracy of traditional dynamic soft sensors may largely deteriorate in complex dynamic processes with nonlinearities, quality constraints or multiple modes. Based on the dynamic latent models, this project intends to provide a soft sensor modeling framework for complicated dynamic industrial processes. First, for nonlinear dynamic processes with small, medium and large sample sizes, soft sensor modeling techniques are researched based on gaussian process, locally weighted learning and deep learning techniques, respectively. Then, supervised and switching dynamic latent models are studied for complicated dynamic processes with different quality constraints and multi-mode dynamic processes, respectively. The proposed techniques will be applied to the industrial hydrocracking processes for prediction performance validation. The research results of this project can provide important theoretic and technical supports for intelligent optimization manufacturing.

软测量可实现工业过程关键产品质量和参数的预测估计，为过程监测、控制和优化提供重要依据。现有静态建模方法无法描述工业过程动态特性和正确的输入输出关系，而传统动态软测量方法描述非线性、质量约束性和多工况性等复杂条件能力受限，造成模型预测精度低和鲁棒性差等问题。为此，申请者在前期研究基础上，提出采用动态隐变量模型，开展复杂动态过程的软测量建模方法与应用研究。首先引入高斯过程描述、局部加权和深度学习技术，分别研究小、中、大三种样本规模下单工况非线性动态过程软测量方法；在此基础上，针对动态过程的质量约束性和多工况切换性问题，分别研究基于有监督动态隐变量模型和多模型切换动态隐变量模型的软测量方法；形成一套基于概率动态隐变量模型的工业过程动态软测量建模体系，并在加氢裂化过程中应用验证。研究成果对丰富软测量建模理论，推进石油化工等典型生产过程的智能优化制造具有重要意义。

在工业软测量建模中，现有静态建模方法无法描述工业过程动态特性和正确的输入输出关系，而传统动态软测量方法描述非线性、质量约束性和多工况性等复杂条件能力受限，造成模型预测精度低和鲁棒性差等问题。为此，项目研究了基于动态隐变量模型的复杂动态过程的软测量建模方法与应用。首先，引入局部加权和深度学习等技术，分别研究不同样本规模下单工况非线性动态过程软测量方法；在此基础上，针对动态过程的质量约束性问题，研究了基于有监督动态隐变量模型的建模方法；最后，针对多工况切换条件下的动态建模问题，研究了注意力机制的自适应动态隐变量建模方法；形成了一套基于动态隐变量模型的工业过程动态软测量建模体系，并在加氢裂化过程和烧结过程等工业现场中进行了应用验证。研究成果对丰富软测量建模理论，为过程监测、控制和优化提供重要依据，推进石油化工等典型生产过程的智能优化制造具有重要意义。