动态偏最小二乘在软测量建模方法中的应用

Application of Dynamic Partial Least Squares in Soft Sensing Modeling

在工业生产过程中,一些重要的过程变量由于技术或经济等方面的原因无法在线测量得到。软测量技术将自动控制理论与生产过程知识有机结合,通过构建数学模型,基于一些容易测量的变量间接获得无法直接测量的变量。软测量技术不仅为关键过程变量的预测提供了一种实时测量的方法,而且有助于优化控制。动态偏最小二乘法(DPLS)以其优良的性能,被广泛应用于软测量建模。对基于动态偏最小二乘的软测量技术进行详细研究。针对将传统动态偏最小二乘软测量方法应用于传输延迟大的系统模型中存在的局限性,采用降维的方法加以解决。在合成橡胶生产过程中,转化率是表征聚合反应进程的一项重要的质量指标参数。通过对丁苯橡胶聚合转化率的仿真表明,该动态偏最小二乘方法可以为存在复杂时变的系统提供更优的关键过程变量预估。

In the process of industrial production,there are some important process variables that cannot be measured online because of technical or economic reasons. Soft sensing technology combines the automatic control theory with the knowledge of the production process. By building mathematical model,based on some easily measurable variables,those variables that cannot be directly measured can be indirectly obtained. Soft sensing technology not only provides a real-time measurement method for the prediction of key process variables,but also helps to optimize the control. Dynamic partial least squares（ DPLS） is widely used in soft sensing modeling because of its excellent performance. The soft sensing technology based on dynamic partial least squares is studied in detail. Aiming at the limitation of applying traditional DPLS soft measurement in the model of large transmission delay,a reduced dimension method is adopted to solve the problem. In the process of synthetic rubber production,conversion rate is an important parameter of quality index that characterizing the process of polymerization. The simulation of the polymerization conversion rate of styrene butadiene rubber shows that this dynamic partial least square method can provide better predictive estimation of key process variables for systems with complex time-varying.