Oil stabilization is a particular application of model-predictive control much like in refining. In this case the oil stabilization unit operation is much like a distillation column where the feed (oil with dissolved gas) comes in approximately in the center of the tower and the tower separates out the dissolved gas. The gas, with some entrained condensate, goes out the top and de-gassed oil goes out the bottoms. Most oil stabilization towers have recycle / reflux loops at the top and bottom of the tower.
In Bintulu, Sarawak, Malaysia, we applied model predictive control to four oil stabilization units (towers). These towers are fed "fizzy" oil with a lot of dissolved gas and they remove just the right amount of gas from the oil. Not too much, not too little. In order to have precise control of RVP, the MLNG gas plant, one of the largest gas processing facilities in the world, wanted a setpoint on "RVP" (Reid Vapor Pressure), the "fizziness" of the oil. They had an RVP instrument on each of the oil stabilization units, but those instruments were substantially downstream on the oil product and thus reported RVP about 20-30 minutes after the fact.
We modeled the current and past values of unit process conditions; temperature, rates and pressures versus what RVP would be in 20-30 minutes, thus the models were predictive, forward looking. The inputs to the models were the current values of operating conditions and the output was RVP in 20-30 minutes. These models were then put on-line as virtual sensors of RVP. Once proven to be quite good, the models were put into our iImprove model-based optimizer tool and the resulting optimization solution was put on-line. This optimizer took uncontrollable factors (feed rate, pressures, etc.) and searched the model for proper temperature setpoint to use in order to target RVP at 11.25. Thus, the MLNG plant had their RVP setpoint and it worked very well.
This type of model-predictive control can be used anywhere you have sufficiently good models and need to control a process. It does not have to be oil stabilization per se, but any type of application of this nature. These "data driven" non-linear regression types of predictive control are particularly useful when the theoretical basis for the setpoint vs. the desired result is unknown or just too difficult to determine theoretically. Examples of such cases include:
This oil stabilization application uses a form of model-predictive control.
The models are created from operating conditions vs Reid Vapor Pressure.
Model-based optimizers are used which determine temperature setpoints to control RVP to setpoint.
This technical capability is useful in many applications in all markets.
Used by major oil and gas companies, such as Shell.
Call: 1-281-760-4007 or send an email.
Do you have a need? If yes, send an email to us and we can discuss.