Measurements in Optimization Models

Online optimizer models use different types of measurements, like pressure, temperature, composition, valve position, flow to adjust internal parameters in the first case, where the objective is to update the model with present plant conditions. These parameters are kept constant during the second case where the optimum conditions are sought to maximize an objective function, it means that new plant conditions are predicted by the model.

The quality of the measurements defines the quality of the parameters and consequently, the optimizer model prediction. Even having a good measurement maintenance system, they can fail; and it is even worse if the quality of the measurement is not well supported.

During model execution, it is important to detect the most important parameters, those that ensure model predicts plant changes with acceptable performance. Then, model is designed including alternative measurements to keep it well-adjusted if the primary measure fails.

It is also necessary to incorporate a method to detect "online" when a measurement fails. There should be different ways to capture the scenarios, sometimes instrument quality detector is not enough, or validity range continues being correct although is not good. A continued tracking and monitoring of measurements quality helps to identify model opportunities and gaps.

Finally, if model is running online and an important measurement fails, a code identifies the situation and switches the model to use the secondary measurement to adjust the parameter. Making a good tuning of that, model will remain reasonably well parameterized and the optimal condition will be predicted well.