Study On Data Rectification For The Steady-State Chemical Process

In this thesis, traditional data reconciliation and gross errors detection methods are studied firstly. Aiming at the difficulty associating with the construction of projection matrix in traditional data reconciliation techniques, a new approach to reconcile process data is proposed, which makes data reconciliation and variables classification not depending on the projection matrix technique. The results of two cases show that the approach is very efficient and have some practicality.Data reconciliation and variable classification of the nonlinear process system usually involve complicated programming. Up to now, a satisfactory schema to solve this problem has not come up. So, a method for data reconciliation based on MATLAB optimization toolbox is proposed, which utilize the successive quadratic programming to solve the reconciliation problem. The efficiency and practicality are demonstrated by several calculation examples.After some defects in traditional gross errors detection and identification strategy based on serial elimination and serial compensation are analyzed, a method which incorporates the linear combination technique and gross errors simultaneous estimation strategy is introduced. Using linear combination technique to determine the subset of candidate gross errors can reduce the size of the combination tests largely and as a result, the time cost by detection and identification is cut down greatly which was displayed by the data rectification for a methanol synthesis unit.According to the fact that the equivalent gross error sets widely occur in the data rectification, a imbalance correlation strategy is carried out to locate gross errors in the equivalent gross error sets accurately. The efficiency of this strategy is illustrated by the identification of measurement bias for a recycle network.