【作者】 李津蓉；

【导师】 戴连奎；

【作者基本信息】 浙江大学 ， 模式识别与智能系统， 2003， 硕士

【摘要】 生产过程中的测量数据是许多技术工作的基础和出发点，它的可靠性和正确性直接影响着研究和决策工作的质量。但由于测量中不可避免的误差，测量值不能精确地符合生产过程中一些内在的物理和化学规律，如化学反应计量关系、物料平衡和热量平衡关系等等。这种误差又分为随机误差和显著误差两大类。数据校正的目的就是综合应用统计、辨识和优化技术，对实测数据进行调整，消除数据中包含的随机误差和显著误差，去掉明显错误的或不可靠的测量数据，从而提高测量数据的质量。 通过对以往数据协调和显著误差检测方法的分析研究，同时将理论研究同生产过程中的实际情况相结合，本文对以前的一些数据校正技术在实际情况中碰到的若干具体问题进行了剖析，并提出来相应的解决方案。具体包括以下几个方面： 1．通过大量的中外文献阅读，对数据协调及显著误差检测技术的发展和研究方向做了一个较为系统完整的阐述。 2．通过构造一个基于测量值比例关系的F统计量，并与约束残差统计量相结合，对稳态过程中出现的显著误差进行检测。这种方法即避免了基于测量残差的检测方法会将显著误差分散到各个测量值中去的缺陷；又避免了基于约束残差的检测方法只能对节点的平衡性进行判断，而无法确定显著误差的具体发生位置的缺陷。仿真研究结果表明：此方法对显著误差十分敏感，其各项性能指标均符合实际工业要求，具有较高的可信度和可应用性。 3．针对传统数据协调模型的缺陷，通过添加一组基于测量值比例关系上下限的约束条件，并利用罚函数的概念将物料平衡的约束条件以软约束的形式表示，由此建立了一种新的数据协调模型。改进后的数据协调模型只会对含有显著误差的测量值给予较大的协调量，而使得显著误差对其他测量值协调结果的影响较小，具有较高的鲁棒性。仿真试验证明：基于该改进模型的协调结果，可直接利用测量残差检测法进行显著误差检测，具有较高的错误检出率，且“虚警”的错误率较低。 4．详细叙述了上述改进数据协调模型在炼油厂连续催化重整装置中的实际应用情况。实际结果表明：本文所提出的数据协调和显著误差检测技术具有显著的应用价值。更多还原

【Abstract】 Measurement data from industrial processes are the basis of much technical work, and data reliability and accuracy will directly affect the results of research and decision-making. However, the original measurement data often contain various errors, so the basic balanced relations, such as energy balance or conservation of mass, cannot be satisfied. Measurement errors can be mainly divided into two types: random error and gross error. Data rectification is a modern technique to improve the quality of measurement data, and its main purpose is to eliminate the random errors and gross errors included in original data by making use of applied statistics, identification, optimization and other techniques.Base on the existing techniques of data reconciliation and gross error detection, this thesis presented some new problems from real industrial processes and proposed the corresponding schemes. The main contributions include:1. Review the development and the state-of-the-art of the techniques in data reconciliation and gross error detection2. For the existing test based on measurement residual, the main problem is that the data reconciliation procedure tends to spread the gross errors overall the measurements, which results in the detection failure. For the methods based on constraint residual, they can be used to only tell which node is imbalance but cannot identify where the gross error is. In order to avoid these problems, this thesis proposed a new test method. The new method combines an F-statistic with constraint residual statistic to detect gross errors in steady state processes. Simulation results show that the new method is very sensitive to the presence of gross errors and has a great probability of correctly finding one or several gross errors.3. In order to avoid the drawback of the traditional data reconciliation model, an improved model is proposed in this thesis. Some new constraints for the ratio of measurement data are added to the new model, and the constraints of mass balance are transformed into soft constraints by using the method of penalty function. The data reconciliation procedure based on the improved model tends to make the measurements having gross errors get more modification than the others. Therefore, the new data reconciliation model is much more robust than the traditional. Besides, the results of the newmodel can be used to detect gross errors directly. Simulation results show that the gross error detection based on the new model is very sensitive to the presence of gross errors.4. The above improved model has been applied to practical data reconciliation for a continuous catalytic reforming unit in an oil refinery. Application results show that the improved data reconciliation model is very effective and can be widely used in industrial processes.更多还原