A Data Reconciliation Approach for Pulp and Paper Simulations: An Illustrative Example, 2003 Fall Technical Conference

Measurement data from the process are sometimes numerous, redundant and unreliable. As a consequence, it is generally impossible to come up with a mathematically sound mass-energy balance using such measured data. On the other hand, a simulation of the same process, which uses first principle mass-energy equations provides values that are by definition balanced values, but are of course different from the measurements on the same variables. Since a simulation is always only an approximation of the real process, and since measurements are always only an approximation of the underlying balanced values, an important question that arises is therefore: how can one utilize measured values to develop a reliable and representative simulation? The answer to this question is the foundation for the important topic called data reconciliation. In this paper, we present a simple data reconciliation strategy that is based on a simplex minimization of the sum of weighted normalized absolute errors between measured and simulated variables, hence yielding reconciled values. This optimization is performed within a commercial simulator which task is to compute balanced values that are as close as possible to the unbalanced measurements. The weighting aspect enables the user to provide levels of confidence between various measurements, i.e. the user can specify that some measurements are likely to be more reliable than others. As such, for the same process, different sets of reconciled data can be found according to the weights assigned to each measurement. We will present example simulations in which the proposed data reconciliation strategy has been implemented.