Data Reconciliation And Parameter Estimation For Process System

The economic performance of real-time optimization system is influenced by the accuracy of the model. If the process model is not consistent with the process plant, it would lead to the offset between the true plant optimum and the predicted optimum and decrease the efficiency of real-time optimization. So it is important to use data reconciliation and parameter estimation (DRPE) to make sure that the process model is consistent with the true plant. The process data measured from the process plant have both random and gross errors. Those errors influence the accuracy of the estimated parameters. It is necessary to use data reconciliation and gross error detection for tuning the measured data and decreasing the effect of random and gross errors. Therefore, data reconciliation and parameter estimation are key parts in real-time optimization system. Some important problems and corresponding solutions in data reconciliation and parameter estimation are proposed in this dissertation. The process data measured from the real plant are used in some DRPE problems. This dissertation makes progresses in the following aspects:1. The definition of parameter estimability is proposed in a system for the DRPE problem. The necessary conditions for parameter estimability are stated in both linear and nonlinear systems. The relationship between parameter estimability and gross error identifiability is analyzed. Finally, the relationship between parameter estimability and the number of data sets is considered.2. A framework for multi-layer data reconciliation in process system is presented, which can adjust the process model to be used for data reconciliation. The characteristics of each layer for data reconciliation are proposed and gross error identifiability in each layer is analyzed. The effectiveness of this framework is demonstrated on the simulation of distillation column system and air separation system.3. The problem of data reconciliation and gross error detection based on the maximum-entropy estimate is formulated. Probability distribution of measurement error is derived from the theory of maximum-entropy estimate and the formulation of data reconciliation is based on maximum likelihood. Additionally, gross error detection criterion is proposed. The effectiveness of the method can be demonstrated by the results of numerical simulations.4. A robust estimator, quasi-weighted least squares, is proposed for data reconciliation. This estimator decreases the effect of gross errors on the objective, and can give accurate results even if the measured data have gross errors. The properties of the estimator are analyzed, and the influence function is used to show that the estimator is robust. Two estimators, weighted least squares and quasi-weighted least squares, are used in atmospheric tower, ethylene separation and air separation process systems. Comparisons with other approaches are made on the steam metering process. The effectiveness of this robust estimator is demonstrated.5. The methodology of data reconciliation and parameter estimation has been presented according to the characteristics and problems in variable load process system. And then, the methodology, which contains steady states detection and sampling, multiple operation conditions clustering, multiple data sets for data reconciliation and parameter estimation, are proposed and used in air separation process. The effectiveness of the proposed methodology can be demonstrated by the results of numerical experiments.6. According to the characteristics of the DRPE optimization problem with multiple data sets, we construct a series of sub-problem based on objective and model parameters. Sequential sub-problem programming strategies for data reconciliation and parameter estimation with multiple data sets are proposed. The solutions of each sub-problem are good initial values for the optimum of the next sub-problem. By solving the series of sub-problem, the optimum of the original DRPE optimization problem can be derived efficiently. The proposed sequential sub-problem programming strategies are used in the industrial purified terephthalic acid (PTA) oxidation process system. The effectiveness of the proposed strategies can be demonstrated by the results of numerical experiments.7. An efficient optimization method based on mnemonic enhancement is proposed for solving large scale and nonlinear problem of data reconciliation. This method uses the experience of pervious solutions to improve the convergence of solution. The frame of this method is designed, and the process of implementation is presented. With the simulations of multi-columns system and ethylene separation process system, the effectiveness of this method is demonstrated.The dissertation is concluded with a summary and prospect of future work in data reconciliation and parameter estimation researches.