半再生催化系统反应性能实时监控、评估与优化

The advanced control theory and methods for real-time monitoring, evaluation and optimization of the performance are presented for the semi-regenerative catalytic systems in the project. The objective of the project is to keep the catalytic system safe and the operation stable and to improve the yield of the desired product. The hybrid statistical model with MIMO (Multi-input Multi-output) based on relative space transformation PLS (Partial Least Squares) and LS-SVM (Least Squares Support Vector Machines) was developed to evaluate the system's performance at first. The principal component extraction method of multi-varibles and information integration were studied. Significant effects of the process variables on the performance were analyzed and their coupling effect mechanism was also investigated, thus the dynamic change laws of the performance were eventually explored. Secondly, the strategy of modeling samples selective sparsity combined projection method and fast leave-one-out (FLOO) ctiterion and the corresponding recursive algorithms were proposed to improve real-time computing performance and generalization ability of the monitoring model. Meanwhile the rapid identification and the early warning system of the abnormal operating conditions were then presented. Finally, The system's performance was optimized by a new hybrid intelligent algrithm based on qualitative analysis and the adjustment strategy was proposed for the operating conditions as well. Hence, this is a project with a very important scientific significance and practical value.

本课题研究半再生催化系统反应性能实时监控、评估与优化的先进控制理论与方法，以保障催化反应系统安全、稳定运行，提高目标产品的收率。首先开发基于相对空间变换PLS和LS-SVM的MIMO混合统计模型评估系统的反应性能，研究多变量主成分提取和信息综合的方法，分析过程变量对反应性能的显著影响关系，查明多变量耦合作用机制，揭示反应性能的动态变化规律；然后提出结合投影技术和快速留一剪切算法的建模样本选择性稀疏策略及相应的递推算法，提高监控模型的实时计算性能和泛化能力，提出异常工况的快速识别方法与预警机制；最后基于定性分析，采用一种新的混合智能算法优化反应性能，提出操作条件的调整策略。因此，本项目具有非常重要的科学研究意义与实际应用价值。

本项目的研究来源于实际工业过程面临的难题，应用系统机理分析方法、统计学习理论，通过理论分析、计算模拟和现场数据验证相结合的途径，研究半再生催化过程的反应性能，揭示多元变量耦合作用下的反应性能的动态变化规律。在研究的过程中，提出了一种基于模糊曲线加权的MMMD数据过失误差侦测方法，提高了过失误差侦测的导向性和合理性。提出了建模样本的选择性稀疏策略，构建了SR-PLS软测量模型，降低了PLS软测量模型的校正频率。开发了基于模型性能评估的双重校正技术，利用双重校正策略分别解决过程模型发生偏移和随机噪声影响下的过程输出修正问题，分析了过程模型通道结构的不匹配问题，给出了评估过程模型的CUSUM技术，开发了局部最小二乘数据拟合方法，给出了偏差校正方法。针对单个模型难以精确描述复杂非线性对象而多模型又多采用静态模型因而对系统实际运行中的动态变化考虑不足的问题，提出了一种基于多模型动态融合的自适应软测量建模方法。针对一类非线性系统的控制问题，结合基于仿射传播聚类的最小二乘支持向量机（LS-SVM）多模型建模算法与PSO优化算法，提出新的多模型预测控制算法。项目研究的成果对半再生催化过程的建模和先进控制具有重要的理论意义和应用价值。共发表论文9篇，其中SCI收录2篇，EI收录7篇。

内容获取失败，请点击重试