一种求解数据校正问题的基于经验增强的方法

摘要

基于机理模型的数据校正问题中,过程系统比较复杂时就要面临求解大规模非线性规划问题.如果直接求解,由于方程维数大且为非线性,问题自由度大,因此求解难度较高,容易导致求解收敛失败.数据校正问题的特点是当测量变量的测量值发生变化时对同一过程对象模型重复地进行求解计算.基于此特点,今提出了一种基于经验增强的求解方法.此方法通过合理地利用以前求解的经验,以达到提高收敛性的效果.设计了此方法的框架及其具体实现步骤,并应用脱丙烷塔和脱丁烷塔的联塔系统与乙烯分离系统进行测试,结果显示相比于传统求解方法,此方法具有很好的收敛性.%Data reconciliation problem based on rigorous model of a complex process system is a large scale nonlinear programming (NLP) problem with large degrees of freedom. NLP solvers could be inefficient to solve this kind of problems and sometimes hard to converge. In view of that the data reconciliation problem should be solved repeatedly based on the same model when the feed condition changes, a mnemonic enhancement based method was proposed to solve the problem efficiently. The proposed method uses the experience of pervious solutions to improve convergence. The frame of this method was designed and the procedure of implementation was presented. The application of this method to simulate the multi-column system and ethylene separation process system demonstrates the effectiveness of this method.