论文部分内容阅读
随着进料负荷、产品组分等过程参数的改变,生产过程的工况也随之改变,而传统的基于多元统计过程监控方法都是假设过程处于单一工况下,因此,将传统方法应用到多工况过程时,往往不能获得很好的监控效果。在对复杂系统进行监控时,由于过程的非线性,传统的基于线性模型的监控方法由于忽略了系统的非线性特征,监控性能也大打折扣。本文针对工业过程中的多工况和非线性监控问题,提出了一种基于即时学习(Lazy Learning)和Greedy-SVDD的多工况过程监控方法。首先使用Lazy Learning对过程进行多模型局部建模,获得局部模型输出和过程真实输出的残差,通过对残差进行监控从而去除多工况的影响。然后用能够很好地处理非线性问题的支持向量数据描述(SVDD)方法对残差建立过程监控模型。为了解决SVDD方法用大样本建模时计算复杂度非常高的问题,本文用Greedy方法提取建模数据集的特征样本用于SVDD建模。最后将上述方法应用在TE模型和乙烯裂解炉的监控上,仿真结果证明了该方法的有效性。
With the change of process parameters such as feed load and product composition, the working conditions of the manufacturing process also change. However, the traditional method based on multivariate statistical process monitoring assumes that the process is under a single working condition. Therefore, the traditional method To the multi-process process, often can not get good monitoring results. In the monitoring of complex systems, due to the nonlinear process, the traditional linear model-based monitoring method greatly reduces the monitoring performance due to ignoring the nonlinear characteristics of the system. In this paper, we propose a multi-process process monitoring method based on Lazy Learning and Greedy-SVDD for multi-process and non-linear monitoring in industrial processes. First of all, we use Lazy Learning to model the process of multi-model locally, obtain the residuals of local model output and process real output, and monitor the residual to remove the influence of multi-conditions. The process monitoring model of the residuals is then established using the Support Vector Data Description (SVDD) method, which deals well with nonlinear problems. In order to solve the problem that the computational complexity of SVDD method is very high when modeling large samples, we use Greedy method to extract the characteristic samples of modeling datasets for SVDD modeling. Finally, the above method is applied to the monitoring of TE model and ethylene cracking furnace. Simulation results show the effectiveness of the method.