Model Predictive control to maximise oil export at Ivar Aasen

Digitalisation through Advanced Process Control

Ivar Aasen exports oil to Edvard Grieg and needs to meet an absolute maximum constraint due to the oil export. With a simple level controller, which controls the oil level in the separator through an oil export pump, the oil level typically varies around a chosen set point. The result of this is variation in the speed of the oil export pump and thus variations in oil export to Edvard Grieg, which makes it challenging to export oil close to the maximum constraint.

The goal of the Cybernetica CENIT NMPC application implemented at Ivar Aasen, is to maximize the oil export rate from the inlet separator, as a result from minimized fluctuations in the flow rate. This is achieved by introducing production swing wells (chokes) for level control of the separator, in combination with a “floating” oil level, instead of a PID level controller.

One of the production chokes is operated automatically, instead of manually by the control room operator, to keep the oil level within predefined minimum and maximum levels in the separator. In addition, the controller may manipulate the pump speed of the oil export pumps, if necessary.

The level deviation from the desired average oil level setpoint in the separator may be larger than what is observed prior to installation of the NMPC application, as the application will utilize the separator buffer volume to a higher extent. The variation in the pump speed is less, close to zero, and thus the variation of the oil flow from the export pumps is minimum, which makes it possible to increase the oil flow export to a maximum without exceeding the maximum export constraint for oil export.

Control philosophy for the Oil Export NMPC implementation at Ivar Aasen

A simple illustration of the control philosophy is shown in the figure blow. The CENIT NMPC application is manipulating the production chokes and the oil pump (green arrows). Only one production choke is manipulated at a time together with the pump. The operator selects min- and max values for the different chokes, in addition to the active swing choke currently used. The operator also defines the target oil level and the corresponding min- and max values in the separator. The NMPC application should fulfil the target oil level on average and allow the level to drift between the limits. The main level control is achieved by manipulation of the swing choke. In addition, the NMPC may manipulate the speed of the oil pump, if necessary. However, the pump speed will mainly be used to achieve the operator specified production rate. In normal situations, the pump speed is thus close to constant.

Figure: Illustration of control philosophy for the IAA NMPC application for oil export

The operator can switch between the NMPC application and the existing level controller (LIC) for oil export as preferred from the HMI. The LIC then starts operating from the last NMPC calculated pump speed value. The swing choke position remains at the last calculated NMPC value, until the operator chooses any manual adjustment.

Using this NMPC application, Ivar Aasen is able to export oil at maximum rate to Edvard Grieg without violating the maximum constraint for oil export, while managing operational changes like flow compositions and disturbances. This resulting in Ivar Aasen being able to export up to 55 000 more barrels of oil per year.


(IAA photo by Aker BP)

Pressure relief system sizing

Problems investigated

  • Sizing of pressure relief valve
  • Safety system constraints on the process operation
  • Safety constraints included in NMPC


  • The Divided Difference Kalman Filter updates the model from a logged data series.
  • Errors are introduced in the model, and predictions are made without updating from that point forward.

Application to vapour pressure systems

An exothermic reaction may exhibit an accelerating reaction rate if the cooling is insufficient. This is referred to as a “runaway reaction”. The runaway causes the temperature to rise, and possibly also the pressure.

A special case is referred to as “vapour pressure systems”, that is when the pressure generated by a runaway reaction is entirely due to the vapour pressure of the reacting mixture, which rises as the temperature of the mixture increases during a thermal runaway. For vapour pressure systems, the emergency relief system is designed so that the action of the pressure relief system removes vapour (and therefore latent heat) at a rate fast enough to hold the temperature  and the pressure constant.

In practice, the pressure relief system is designed so that the maximum runaway pressure is less than 110 % of the opening pressure of the relief valve.


The scenario typically studied, is loss of cooling at the worst possible time during an exothermic reaction.

For a batch polymerization system,  several studies indicate that loss of reactor cooling is more critical in the middle of the batch than at the beginning and the end.

Another important and related result is that traditional “self heat rate” experiments and calculations, tend not to capture the worst case scenario and may lead to under-dimensioned safety system capacity.

Aluminium electrolysis

The aluminium production is a high-temperature process with a chemically aggressive bath. The harsh process conditions means that measuring the internal process states is a formidable challenge. Typically measurements are performed infrequently on key variables. Instead, more peripheral but technically feasible measurements are monitored continuosly. This can typically be temperature measurements, offgas analysis and electrical measurements. The challenge of infrequent and poorly available measurements is typical for many metallurgical processes.

  • By the use of an online process model and estimator, Cybernetica provides a solution for soft-sensor measurements, filling the time gaps between the physical measurements.
  • The process model calculates real-time variables that are normally not available as measurements.
  • The augmented process information provides insight into the process which enables model-based solutions.

Cybernetica is a partner in research and development projects together with industrial partners, research institutes and universities, as well as other specialized technology companies. Research and development project are often partly financed by the Research Council of Norway or the European Union.

Key working areas and solutions:

  • Training simulator and educational gaming.
  • Real-time monitoring of process variables.
  • Controlling the process using model-based predictions.

Metal refining processes

In some refining processes, this could be done by batch-to-batch optimizing of fines additions and optimized flow profiles of refining gases. In batch refining processes where some chemical analyses and measurements are available during the refining, a full Model Predictive Controller (MPC) can be implemented as a monitoring and operation support system, as well as closed loop control.
End point prediction of product quality and production yields, as well as optimized economical margins are key variables for the MPC to calculate and optimize on.

During the life time of the refining equipment (ladles, converters, gas nozzles, etc) are worn, or material grow onto them, causing the refining to become less optimal. These changes are often slowly evolving until the equipment is so worn or damaged, that it needs to be replaced. Adaptive techniques using estimators are used to identify this change in behviour.  Having updated parameters in the model, ensures that the optimization is still valid even if the equipment characteristics may change.

Electric arc smelting furnaces

However, by using model calculation in combination with available process measurements and analysis, information about the process that is otherwise unavailable, can be obtained. An example of process information otherwise unavailable is the temperature and chemical profile of the charge burden from the furnace top, down to the electrode tip.

Cybernetica has long experience in modeling of electric arc smelting furnaces for the ferroalloy industry. Cybernetica’s process models are typically based on first principles, with mass and energy balances, describing physical phenomena in the processes. Modeling of an electric arc smelting furnace typically includes both a metallurgical model of the chemistry and material properties as well as an electrical model.

Cybernetica also has experience from modeling of submerged arc furnace processes such as ferrosilicon/silicon, ferromanganese, silicomanganese, and slag smelting processes.

Handling of liquid slugs – Avoid reduction in production

Coordinated separator control

In a general case, the setpoints of all LIC controllers, marked with green arrows, might be manipulated by the model predictive controller. The number of manipulated variables is decided from operational requirements.

Multivariable control of a separator train

The red markers indicate possible controlled variables. As indicated, both level values and flow (calculated or measured) of oil and water out of each separator may be controlled explicitly. It may seem a bit strange that the MPC controller includes level control, as this is the primary task for each LIC. However, the individual LIC controllers are monovariable controller loops, only considering a single level transmitter and a single valve. When introducing multivariable control, we choose to keep the basic PI(D) controller loops for safety reasons, and the setpoint of the LIC is thus only an indirect way of manipulating the liquid flow valves. The basic controller loops, including the PI(D) tuning, is therefore included in the prediction models of the MPC application. This is similar to how an experienced operator may act if he becomes aware of an incoming liquid slug: He decreases the LIC setpoint in order to drain the separator, not because he wants a lower level, but because he realize that without his intervention, the LIC controller alone, with a fixed setpoint, will give an unacceptable high level.

An MPC application can handle multivariable level control, include prioritazions between different levels, and also handle max- and min-values together with setpoint values in a consistent way.

Scenario 1: A slug is estimated to enter first stage:

The first stage separator is probably the largest liquid volume, and the MPC application will challenge the high- and low levels by manipulating the LIC setpoints in order to “filter out” the flow pulse to the second stage or to the water treatment system. If the level predictions for the second stage reveal safe operation of this separator, also the level limits here may be utilized to further damp the flow to the third stage. If necessary, even the LIC setpoints of oil and water in the third stage may be manipulated accordingly to avoid level alarms, or even process trips.

Scenario 2: A slug is estimated to enter second stage only, from a satellite well:

The figure indicates a possible multiphase source entering between first and second stage, for example from a satellite field requiring a lower inlet separator pressure. If this particular flow is sluggish, we may also in this case use the first stage separator as a “compensator”: The separator LIC loops can be operated so that the outlet liquid flow oscillates with opposite phase compared to the satellite inlet flow. The second stage separator may then experience close to a non-slug behavior of the inlet flows. If necessary, we may still operate the second and third stage level controllers as explained in scenario 1.

Finally, if there is a distillation column downstream the separator train, we may regard the flow out of the third separator as a main variable for reduced oscillations.

Norne FPSO, photo by statoil, Anne-Mette Fjærli

Pressure and pipeline control – increased oil production

Gas capacity control

With this controller proposal, we assume that total production capacity is limited by the gas capacity, or more specifically the compressor capacity. As the inlet separator pressure is usually controlled by a standard PIC that manipulates the compressor speed, this means that the speed (or turbine efficiency) has reached its maximum, giving a separator pressure on or above the setpoint value of the PIC. In order to prevent pressures close to the flare limit, the operator may choose to back off production form certain wells or well clusters. However, backing off represents production decrease, and reducing too much will cause the turbine to decrease speed or efficiency, and the plant is no longer constrained by the compressing capacity.

In order to limit the production reduction, and in order to maintain the highest possible pressure at all times, Cybernetica propose a model based controller. The figure below indicates possible manipulated variables with green arrows, and possible controlled variables with red markers. This specific plant configuration indicates subsea wells routed to a common template, in addition to “disturbances” from other wells (“Disturbance” block), whose details are not shown. As a minimum controller configuration for this specific plant example, we should select at least one subsea choke (swing producer) together with the topside choke as manipulated variables. We are then able to control both the separator pressure and the common template (flowline) pressure. The need for inlet separator pressure control is explained above, and the template pressure is included in the MPC controller in order to keep fixed downstream conditions for the non-used common template wells.

Example of operation:

Imagine a reduction in flow from one or more “Disturbance” wells. This will reduce the separator pressure, and actions must be taken in order to remain on the maximum capacity constraints of the gas processing train. Assume that one subsea choke (e.g. the leftmost of the four chokes marked in the picture) can be manipulated automatically in addition to the cluster topside choke. The MPC controller predicts the response of the “Disturbance” wells, and compensates by opening the topside choke for the selected swing cluster. This will obviously increase the pressure of the separator, but it will also decrease the template pressure. We want to avoid the template pressure decrease because of the remaining cluster wells. The controller should therefore also open the selected (leftmost) subsea choke to compensate for the topside choke. However, using the subsea choke will also influence on the inlet separator pressure. We therefore realize that the 2×2 controlling scheme has strong interactions, which motivates for model based control.

Bilde tekst?
Multivariable control of template- and inlet pressure

Controller extensions

The above example could be extended by including several subsea chokes in the MPC application. In the example figure, we have marked all four subsea chokes as possible manipulated variables. We then get more manipulated variables than setpoint controlled variables, which can be handled by some kind of priority hierarchy for the selected chokes. Alternatively, all subsea chokes except from one can be given ideal resting values (steady state values), and thus only be used dynamically to improve the dynamic response.

Deduction well testing

Subsea wells producing towards a common template or flowline will typically be tested by a deduction methodology. Having installed the MPC application described above, it could even be used for test purposes. The choke for the selected well to be tested will be throttled manually by the operator. The controller should keep the template pressure at setpoint, in order to ensure that oil-, gas- and water flow from the non-tested cluster wells remains constant. This could be done by manipulating the topside flowline choke. In this situation, inlet separator pressure is not considered as a controlled variable, and the controller structure will be a SISO (single input – single output) system. However, “feed-forward” from the choke of the tested subsea well is included.  The concept is illustrated in the figure below.

Automatic control during deduction well test
Automatic control during deduction well test

An alternative approach could be to skip control of template pressure, and obtain non-changed flow conditions for the non-tested wells by controlling their corresponding downhole pressures to fixed values. In this case, all subsea chokes except for from the one for the tested well, will be used by the MPC application.

Troll C, photo by Statoil, Øyvind Hagen

Mature fields – decreasing reservoir pressure

For mature fields, production capacity is restricted by the reservoir pressure, and production- and subsea chokes are typically fully open in order to minimize the flow resistance. If possible, effort should be made in order to minimize the processing pressure, e.g. the inlet separator pressure, while at the same time respecting constraints in the separator- and compressor trains. As the MPC decreases the inlet separator pressure, the (export) compressor speed increases, which both increases the gas flow and decreases the suction pressure. The delivery pressure of the upstream recompressor decreases, and obviously both flows and pressures in the entire separator- and compressor trains are altered. To ensure safe operations of all basic PI(D) controllers during changing feed conditions, and to fulfill the final oil specifications (RVP), a multivariable MPC application is required, which are using several PIC and LIC setpoints as manipulated variables.

For some production facilities, reservoir pressures may have decreased so much that production of liquids is impossible without some kind of artificial lift. Recirculated gas is then injected in the well bottoms in order to reduce the density of the fluids and thus the static pressure drop along the wells. Alternatively, this may be viewed as pressure reduction in the wellbore, in order to increase drawdown.

The total available amount of gas should now be distributed among the producing wells in order to maximize the total flow of oil. An MPC application will calculate this optimal distribution, which also could change dynamically because of time varying GOR and GLR for different wells.


The huge advances in fish farming over the last decades has increased the demand for high quality aquaculture feed. With the increase in knowledge on fish health and nutrition, the aquaculture feed producers experience an increasing demand for tailored feed products. At the same time, technical advances in fish farming, such as bulk transport and automatic feeding, pose higher demands on the mechanical strength of the feed. In order to compete in a competitive market, the aquafeed producers must display flexibility, to meet the customer’s demands, while producing large volumes of high quality feed at a high efficiency.


CO2 capture

Over the years, research on climate change and global environment has led to an increased level of awareness regarding the possible negative consequences of human-made greenhouse gas emissions to the earth atmosphere. Emission of CO2 originating from large industrial point sources (e.g. energy production from combustion of coal, natural gas or bio-fuels, production of materials such as aluminium or cement, etc.) is one portion of the big picture of particular interest. Due to an increasing global demand for energy, consumables and utilities, and a rather slow global introduction/expanding of renewable/environmental friendly energy sources, these industries are not expected to rapidly decrease in extent, but the possibility for removing CO2 from these point emissions, thus significantly reducing the carbon footprint, is gaining increased attention.

When it comes to CO2 capture, CO2 absorption using amine-based solvents is a promising technology. Adding a capture plant to a power plant will constitute a significant power cost due to the regeneration process for the amine solvents, and this loss will strongly depend on whether the plant operates optimally or not. In a hypothetical (but plausible) future scenario where companies are forced (for instance by governments) to perform CO2 capture and/or are penalized for emitting CO2, the entire production plant profitability will depend on the optimal operation of the capture plant.

Cybernetica develops model-based solutions for amine-based CO2 capture plants. Together with Sintef and Technology Center Mongstad, we are aiming to demonstrate how capture plants can operate more effectively using Nonlinear Model Predictive Control (NMPC). The capture plants are complex, interconnected processes with a variety of unit operations. Understanding the behavior of the plant under various operating conditions is crucial in order to achieve optimal operation. A typical challenge for power plants is that the energy demand will govern the power plant load and consequently the emission rate of CO2, to which the capture plant must adapt. This will give rise to diurnal variations in plant inputs, making the operation more challenging. In on-line use, the model-based approaches are intended for monitoring (soft sensing) and control purposes. This will strive to aid the plant operators and ensure optimal operation.

Model predictive control of polyolefin processes

The main benefits of using this technology compared to conventional control are:

  • Improved productivity by operating closer to constraints. 4-7% increase in production rate is often feasible.
  • Smoother, faster and safer grade transitions, resulting in less off-spec material.
  • Consistent control of product quality parameters such as melt index, density (HDPE) and copolymer content (PP).
  • Reduced consumption of catalysts and inerts.
  • Fewer production interruptions.

Process modelling

The mechanistic process model utilized within Cybernetica CENIT is adapted to each specific process unit. Reactor design data are acquired from the customer. Other model parameters are estimated off-line from logged process data and lab measurements using Cybernetica ModelFit.

Models of polyolefin processes

  • Mass, energy and component balances.
  • Kinetics and thermodynamics for the polymerization reactions.
  • Phase equilibria.
  • Polymer properties such as MFR, density and co-monomer content, etc.
  • Catalyst deactivation.

Model predictive control (MPC)

MPC will typically be used to manipulate feed rates of monomers and catalysts, as well as other variables dependent on the particular polymerization process.

The figures below show how the control of H2 concentration, solid concentration and production rate are improved with the use of MPC based on mechanistic models as outlined above.

PP loop reactor. Control of H2 concentration during grade transition.
PP loop reactor. Control of H2 concentration during grade transition.
PP loop reactor. Production rate can be increased by reducing variability.
PP loop reactor. Solid concentration variability is reduced with MPC.

Safe operation

Using model adaption, Cybernetica CENIT ensures that the industrial processes are safely controlled in spite of model uncertainty.

In addition to the inherent safety associated with the model predictive control methodology and the online model adaptation, CENIT includes functionality for fault detection and diagnosis.


Cybernetica CENIT applications typically run on a dedicated application station (Windows server). It communicates with a DCS system via the Open Platform Communications (OPC) protocol.

Cybernetica employees have extensive experience in industrial implementation and commissioning of advanced process control (APC) applications.


Cybernetica’s maintenance programme ensures that all applications perform at their best at all times. Cybernetica rapidly responds to customer requests. Cybernetica CENIT has built-in functionality for reproduction and diagnosis of current or past process situations, facilitating efficient analyses of issues reported by the customer.

The most common and important issues are reported directly to the operators, enabling the operators to act when necessary.