SIMULTANEOUS STRATEGIES FOR DATA RECONCILIATION AND GROSS ERROR-DETECTION OF NONLINEAR-SYSTEMS

Process measurements are taken in chemical plants for the purpose of evaluating process control OT process performance. As a result of random and possibly gross errors, these measurements do not generally satisfy the process constraints. Thus they need to be reconciled. In the data reconciliation process, the data are adjusted to satisfy the process constraints while minimizing the error in the least squares sense, and the unmeasured variables are estimated whenever possible. In this study, we instead minimize an objective function that is constructed using maximum likelihood principle to construct a new distribution function, which takes into account both contributions from random and gross errors. The advantages of minimizing this objective function are that it gives unbiased estimates in the presence of gross errors and that simultaneously a gross-error detection test can be constructed based on their distribution functions without the assumption on the linearity of the constraints. Furthermore, the structure of this objective function can be exploited under certain conditions. Thus, efficient nonlinear programming strategies, similar to the hybrid SQP method introduced by Tjoa and Biegler in 1991 for least squares objective functions, can also be developed. The effectiveness of this strategy is demonstrated on nonlinear example problems.