(210b) Advanced model development from production data using multiple objectives

The development of a model for a production reaction system has numerous challenges. The production process even in the best of situations will not include all measurements that the model will be predicting. There will be noise in the measurements, failure of sensors, and systematic errors. There will also be unmeasured factors that can impact the data. There will also be experimental data and designed data that needs to be incorporated into the model.

The model is usually designed to use some of the measurements as inputs while predicting the other measurements. A combination of transport theory and reaction theory is incorporated into the final model with a number of fitting parameters to correct for the unknowns and the simplifications which have been assumed. Often a weighting is assigned to different measured variables to designate their importance to the predictions. The different fitting parameters are adjusted to minimize the weighted difference between the results and the measurements. These fitted parameters can vary greatly depending on the weighting factors that are used.

In this presentation, we look at addressing the errors in the measurements and adjusting fitting parameters to match multiple results using a variety of optimization techniques. An understanding of the appropriate weightings between the different objective functions is gained by using Pareto front analysis. The fitting process is also used to provide insights into the measurement errors and the problems with some of the underlying assumptions used to build the original model. The model is then used to provide further insights into the production process.