WO2016180968A1

WO2016180968A1 - Control system for controlling a dynamic system

Info

Publication number: WO2016180968A1
Application number: PCT/EP2016/060856
Authority: WO
Inventors: Vinicius DE OLIVEIRA; Sigurd SKOGESTAD; Johannes JÄSCHKE
Original assignee: Norwegian University Of Science And Technology (Ntnu)
Priority date: 2015-05-13
Filing date: 2016-05-13
Publication date: 2016-11-17
Also published as: GB201508125D0

Abstract

A control system for controlling a dynamic system comprises: a control loop incorporating a baseline controller with an adaptive control layer, wherein the baseline controller is for controlling a manipulated variable to adjust a controlled variable in accordance with a setpoint for the controlled variable, and wherein the adaptive control layer is for providing adaptive feedback control, the adaptive control layer being arranged to adjust the control of the manipulated variable in response to changes in the dynamics of the controlled system by adjusting parameters of the baseline controller and/or by adjusting the control input for the manipulated variable; and an autonomous supervisory layer for monitoring performance and stability of the control loop and for adjusting the setpoint, wherein the autonomous supervisory layer is arranged to move the setpoint toward a more stable value when potential instability is identified and to move the setpoint toward a potentially less stable but higher performing value when the control loop is stable, wherein the supervisory layer is arranged to check for oscillations in the system and, if there are no oscillations within set thresholds, to adjust the setpoint toward a higher performing value, wherein the supervisory layer is arranged to identify variations in parameter(s) of the system that indicate a potential problem and to adjust the setpoint toward a more stable value in response to potential problems, and wherein the changes to the setpoint by the supervisory layer provide excitations to the dynamic system to enable the adaptive controller to work effectively.

Description

CONTROL SYSTEM FOR CONTROLLING A DYNAMIC SYSTEM

The present invention relates to a control system for controlling a dynamic system, for example the control system may be for controlling flow to reduce slug flow in a pipeline and riser system of the type used in the oil and gas industry.

When a multiphase fluid comprising gas and liquid flows in a conduit with elevation changes, such as a pipeline for oil and gas incorporating a riser, then undesirable slug flow can occur. Slug flow, or slugging, is when "slugs" of liquid exist between areas of gas or multiphase flow, partially blocking the conduit and allowing pressures to build up in the gas. A typical example occurs when the riser section of a pipeline fills with liquid, preventing gas from flowing out until the pressure in the gas below the riser increases sufficiently to push the blocking liquid and gas up through the riser. The escape of gas typically happens at relatively high pressure expelling liquid and gas at a high rate through the riser in a "blow out". This type of slug flow creates serious issues both upstream and downstream of the area where the slugs have occurred due to oscillations in pressure and in flow rates. In the context of oil and gas production slug flow has the potential to damage equipment and adversely affect production rates.

Consequently, there is a need for an effective solution for suppressing slug flow in this type of pipeline. One way to prevent slug flow behaviour is to operate the system at an increased pressure, and in a typical riser arrangement this can be done by reducing the opening of a topside valve such as a choke valve. This solution has the disadvantage that it reduces the flow rate of the multiphase fluid, which is equivalent to a reduction in the production rate when an oil and gas system is involved. Systems have been developed allowing for active control of a topside choke valve with the aim of providing a non- oscillating flow along with maximum possible production rate. Such systems make use of measurements such as pressure, flow rate and/or fluid density as the controlled variables, with the degree of opening of the topside choke valve being the main manipulated variable. An example of this type of system can be found in US 2010/307598.

For maximum production in an oil and gas installation it is desirable to have the lowest possible pressure (maximum valve opening) in the pipeline/riser system. This translates into low pressures at the bottom of the wells, which maximises the fluid inflow from the reservoir. However, as the pressure set point for the pressure controller is decreased then stabilisation of the system and avoidance of slug flow becomes more difficult. The choice of the ideal set point is therefore a difficult task, and in fact the ideal pressure set point will vary with flow and reservoir conditions. Without an automated system for effectively controlling the degree of opening of the choke valve then there is a need for constant monitoring of the control system by experienced operators. In the prior art mentioned above, US 2010/307598, a plurality of sensors are used to obtain information about fluid flow properties and the measured data is analysed to produce a weighting parameter indicative of the severity the of slug flow. The weighting parameter is used in an automated feedback control of the degree of opening of the choke valve, which has the effect that the check valve opening will oscillate around and tend toward a setpoint. The setpoint is checked periodically and if the system is operating in a stable manner than the choke valve opening setpoint can be increased (increasing the degree of opening of the choke valve and lowering the pressure in the system). There is also a periodic repetition of the data gathering step allowing for the weighting parameter to be adjusted taking account of changes in flow conditions. The proposal in this document hence devises a new and non-standard feedback controller and requires a large amount of sensor data to operate effectively.

Viewed from a first aspect, the invention provides a control system for controlling a dynamic system, the control system comprising: a control loop incorporating a baseline controller with an adaptive control layer, wherein the baseline controller is for controlling a manipulated variable to adjust a controlled variable in accordance with a setpoint for the controlled variable, and wherein the adaptive control layer is for providing adaptive feedback control, the adaptive control layer being arranged to adjust the control of the manipulated variable in response to changes in the dynamics of the controlled system by adjusting parameters of the baseline controller and/or by adjusting the control input for the manipulated variable; and an autonomous supervisory layer for monitoring performance and stability of the control loop and for adjusting the setpoint, wherein the autonomous supervisory layer is arranged to move the setpoint toward a more stable value when potential instability is identified and to move the setpoint toward a potentially less stable but higher performing value when the control loop is stable, wherein the supervisory layer is arranged to check for oscillations in the system and, if there are no oscillations within set thresholds, to adjust the setpoint toward a higher performing value, wherein the supervisory layer is arranged to identify variations in parameter(s) of the system that indicate a potential problem and to adjust the setpoint toward a more stable value in response to potential problems, and wherein the changes to the setpoint by the supervisory layer provide excitations to the dynamic system to enable the adaptive controller to work effectively.

As discussed below with reference to an example implementation relating to slug flow, a synergy arises through the combination of an adaptive control layer and an autonomous supervisory layer. The adaptive control allows the baseline controller to adapt effectively to changes in the setpoint and changes in operating conditions. The supervisory layer enables the setpoint to be adjusted to increase the performance of the system, whilst also reacting to potential instability, and at the same time the changes to the setpoint by the supervisory layer provide the excitations required to enable the adaptive controller to work effectively. It will be appreciated that such a system can be applied to any system where the dynamics change when approaching the (generally unknown) operating limit of the system. Such systems include oil and gas systems with two-phase flow, as well as two- phase refrigeration circuits.

The system has been shown to be of particular benefit where the manipulated variable is a valve opening and the controlled variable is a pressure upstream of the valve. The control system can be used to improve flow characteristics for a multiphase flow through a pipeline, especially where there is an inclined section of pipeline such as a riser with a valve at the top or bottom of the inclined section. The valve hence may be a valve at the top of a riser in a pipeline system for multiphase fluid flow.

In such a system for controlling a valve to improve flow characteristics for multiphase flow the control system will be effective in reducing undesirable slug flow. The supervisory layer can decrease the setpoint to thereby decrease the valve opening and increase pressure when there is a risk of unstable slug flow, and it can increase the setpoint to hence increase the valve opening and decrease the pressure to increase flow through the pipeline when the control loop is stable.

Thus, in a preferred example the control system is for reducing slug flow in a pipeline for multiphase fluid, where the pipeline includes an inclined section such as a riser and a valve that controls flow through the inclined section, for example a valve at the top of or at the bottom of the inclined section, the control system comprising: a control loop incorporating a baseline controller with an adaptive control layer, wherein the baseline controller is for controlling an opening of the valve to adjust a pressure or other

measurement within the pipeline in accordance with a setpoint, and wherein the adaptive control layer is for providing adaptive feedback control, the adaptive control layer being arranged to adjust the control of the valve opening in response to changes in flow conditions within the pipeline; and an autonomous supervisory layer for monitoring performance and stability of the control loop and for adjusting the setpoint, wherein the autonomous supervisory layer is arranged to move the setpoint to a more stable value when potential instability is identified and to move the setpoint to a potentially less stable but higher performing value when the control loop is stable. For example, the baseline controller may be for controlling an opening of the valve to adjust a pressure within the pipeline upstream of the valve in accordance with a pressure setpoint or to adjust the valve opening in accordance with a valve opening setpoint and in this case the autonomous supervisory layer may be arranged to increase the pressure setpoint/decrease the valve opening setpoint when a potential instability is identified and to decrease the pressure setpoint/increase the valve opening setpoint when the control loop is stable. Through the combination of adaptive feedback and an autonomous supervisory layer it is possible to effectively control the valve to achieve the minimum

pressure/maximum valve opening (and hence maximum performance of the system) whilst avoiding the risk of slugging. When the system is operating stably then the supervisory layer will gradually decrease the pressure setpoint or increase the valve opening setpoint until just before the control performance is no longer acceptable due to the risk of slugging. The baseline controller may be a standard linear controller and in fact could in some circumstances be a controller already in place at an installation. The adaptive control layer ensures that there is good performance for all setpoints and during all flow conditions by adjusting the control of the manipulated variable.

Without the adaptive control then large variations in the setpoint or large changes in the flow conditions would result in unacceptable performance. Without the use of the supervisory layer to modify the setpoint then manual monitoring and control would be required to achieve performance gains. Moreover, typical adaptive control processes require ongoing variations in the operating parameters for the controlled and manipulated variables, and advantageously these can be provide by excitations resulting from the periodic changes to the setpoint directed by the supervisory layer. Consequently the adaptive control layer and the supervisory control layer combine synergystically to provide significant improvements in the performance of the system, which are not possible with prior art control systems.

It is an advantage of the above control system that only a single measurement for the controlled variable is required. The system could make use of additional

measurements, but it is not necessary to introduce more than one sensor, and typical systems would already include such a sensor, for example for use with an existing controller (which could form the baseline controller for the above control system). Thus, the control system may use only a single measurement, for example from a single sensor. In examples where the controlled variable is a pressure of the system and the manipulated variable may be a valve opening, then only a single pressure sensor is required, the sensor being for sensing the pressure within the pipeline upstream of the valve. In this case the setpoint may be either a pressure setpoint or it may be a valve opening setpoint. A pressure setpoint has the advantage that there is no need to model or calculate a relationship between the valve opening and the pressure upstream of the valve and hence this could provide greater accuracy, although either possibility can be used and the choice might be determined by the nature of a pre-existing control system that is being updated to include features of the new control system discussed herein.

Each element of the control system (the baseline controller, the adaptive layer and the supervisory layer) may make use of the same measured value relating to the controlled variable, such as the pressure in the above example. Thus, the baseline controller may monitor the measured value and use this to determine adjustments to be made to the manipulated variable, the adaptive control layer may monitor the same measured value and use it to determine adjustments to be made to the parameters for the baseline controller, and the supervisory layer may monitor the same measured value and use it to determine the stability of the control loop and thereby enable a decision to be made about changes to the setpoint. The use of a single pressure sensor for all elements of the control system makes the system simpler and easier to implement with pre-existing systems, such as pipeline/riser systems that are already in place.

The supervisory layer is preferably able to quickly detect problems in the control loop and thus may be arranged to identify variations in parameter(s) of the system that indicate a potential problem, for example a potential instability in the control loop. The supervisory layer is arranged to check for oscillations in the system and, if there are no oscillations within set thresholds, to adjust the setpoint toward a higher performing value, for example to a lesser pressure/larger valve opening. The setpoint may be adjusted by a set step size for example where the setpoint is for a valve opening it may involve increasing the degree of the valve opening by at least or about 1%, 2% or 3%. This might be done by adjusting a pressure setpoint by a lesser value, for example at least or about 0.5%, 1 % or 2%. The pressure setpoint and the degree of valve opening may be related non-linearly. The supervisory layer is arranged to detect potential instability and to adjust the setpoint toward a more stable value, for example to a higher pressure/smaller valve opening.

The instability detection may include any technique to indicate that the controlled variable is diverging from the required setpoint (as opposed to converging toward the setpoint). For example, the supervisory layer may be arranged to evaluate the integrated absolute error for the control loop over a time window and compare it to a pre-specified threshold to assess the stability of the control loop. The pre-specified threshold may be a value set as to provide a clear distinction between a potential problem and normal/expected noise in the system, and hence may thus be set as a multiple of integrated absolute error values that arise with noise, for example a value that is two or more, three or more or four or more times larger. The integrated absolute error values that arise with noise may be determined based on historic data or industry knowledge. Oscillation detection procedures may be used, for example the supervisory layer may be arranged to identify growing oscillations with a frequency in a certain predefined range, typically a frequency considerably lower than the average frequency of adjustment of the manipulated variable by the baseline controller. The frequency in a certain predefined range may be set based on a known, predicted, or modelled slug frequency for the system (or a frequency for a similar instability when the problem being addressed is other than slugging) and it may be set as a multiple of that frequency, for example at least or about 1.5 or two times the slug frequency. A frequency analysis, for example a Fourier analysis, may be used to identify oscillations at frequencies associated with potential problems, for example frequencies known to relate to slug flow when the control system is for a multiphase fluid flow system. Such frequencies might be identified using any suitable approach, for example based on past measurements and experience of the system concerned, based on modelling, or based on simulations.

When the supervisory layer detects a potential instability using features as discussed above then the setpoint is adjusted to a more stable value. The adjustment may be based on a default step adjustment, or it may be a variable adjustment with the size of the adjustment being set based on an assessment of the size of the potential instability, for example based on the integrated absolute error value discussed above. In the case of a valve opening where the setpoint is defined as a percentage representing the degree of opening of the valve, the setpoint may be adjusted by at least or about 2%, 3%, 5% or 10% for example. The setpoint may be a pressure set point determining control of the valve and in this case the setpoint may be adjusted by a lower percentage, resulting in a higher degree of valve movement, for example at least or about 1% or 2%. As noted above there may be a non-linear relationship between a pressure setpoint and the degree of valve opening. The supervisory layer may also direct an adjustment to the adaptation values set by the adaptive control layer, for example a return to the values being used prior to the start of the instability.

The supervisory layer may be arranged to detect the impact of a major disturbance in the controlled system and to quickly adjust the setpoint to avoid high amplitude frequency oscillations caused by too high a control gain and/or quickly developing instabilities when the pressure diverges swiftly from the setpoint, which may occur after a large change in operating conditions. Such problems might not be detected in time by the features discussed above, which are focussed on slowly building instabilities.

One possible way to detect too high a gain is to check for higher frequency oscillations, and or to identify the frequency or frequencies with the highest amplitude(s) and compare against threshold values. In one example the supervisory layer may be arranged to carry out a frequency spectrum analysis in order to identify the frequency or frequencies showing the highest amplitude, and then check the amplitude(s) against a threshold value as well as optionally checking the frequency(/ies) against a threshold. The threshold amplitude may be set based on a multiple of an amplitude of oscillations due to noise, which could be based on a measured, predicted or expected noise value. The supervisory layer may determine that there is a potential instability when there is an oscillation at a frequency over the threshold frequency and with an amplitude above the threshold amplitude value. A check for high amplitudes of the highest frequency(/ies) is an effective check for unusual oscillations. The highest frequency component in a stable system would typically be noise, which would have a relatively low amplitude.

Consequently a combination of high frequency and large amplitude is a clear indicator of something unusual (for example, too high gains in the baseline controller) and a potential problem.

In the example where the system is used to reduce slug formation in a pipeline with multiphase flow then the threshold frequency may be defined with reference to the frequency of slug formation for the system without active control of the valve opening. This frequency may be determined based on past measurement/monitoring of the system or of similar systems, based on modelling, or based on simulation, for example. A typical threshold might be set at 10 times the reference frequency for slug formation.

In order to identify quickly developing instabilities involving large changes in pressure or other system parameters, the supervisory layer may be arranged to check the rate of change of the integrated absolute error for the control loop over a time window, and to identify a problem when this is increasing beyond the threshold values discussed above despite corrections and/or when the rate of change of the integrated absolute error is above a threshold rate. When reacting to quickly developing instabilities, including potential instabilities detected as described above, then the supervisory layer is arranged to adjust the setpoint to a more stable value. The adjustment may be based on a default step adjustment, or it may be a variable adjustment with the size of the adjustment being set based on an assessment of the size of the potential instability, for example based on an amplitude of a high frequency oscillation that is indicative of an unstable control loop. It may be an advantage to adjust the setpoint by a larger value than the adjustment made for instabilities that manifest in a more gradual fashion. In the example of a valve opening with the setpoint defined as a percentage representing the degree of opening of the valve then the setpoint may be adjusted by at least or about 10%, 20%, or 30%, for example. As with the slower developing instabilities, when a quickly developing instability is detected the supervisory layer may also direct an adjustment to the adaptation values set by the adaptive control layer, for example a return to the values being used prior to the start of the instability.

The baseline controller may be provided with an integral adaptive controller, hence taking the form of a fully adaptive controller. However, it is preferred to use a conventional baseline controller and to augment it with an adaptive layer. This enables the control system to be easily added to existing control systems. In this case the baseline controller can be any known controller suitable for use with the controlled system. Preferably a linear feedback controller is used. The addition of the adaptive layer and supervisory layer will provide improvements even with a poorly tuned baseline controller. A well-tuned baseline controller will however provide better performance. The baseline controller can be selected from known feedback control systems, for example proportional integral (PI) controllers, loop transfer recovery (LTR) controllers, Hjnfinity controllers or Model Predictive

Controllers. The baseline controller is preferably arranged to provide a desired

performance and robustness under nominal conditions. It will be appreciated that the adaptive control layer then provides the ability to maintain performance even when operating conditions change since the adaptive control layer can adjust the parameters of the baseline controller accordingly.

The adaptive control layer may be any suitable controller capable of adjusting the control of the manipulated variable in response to changes in the dynamics of the controlled system by adjusting parameters of the baseline controller and/or by adjusting the control input for the manipulated variable. The latter is used in some preferred examples since it provides a simple add-on to an existing baseline controller. The control input to the manipulated variable may be a summation of the output from the baseline controller and an adaptive augmentation control signal from the adaptive control layer. Preferably a robust adaptive control layer is used. The adaptive control layer may be based on a reference model for the controlled system, i.e. a model with similar dynamic characteristics to the controlled system, which may advantageously be an observer-like reference model. This can improve the transient dynamics of the adaptation scheme.

In the example where the controlled system is a pipeline with a riser then the valve may be a choke valve, and it is typically a valve that is already present in the pipeline system. The pipeline system may be a part of an oil and gas installation, for example an oil and gas production installation. One preferred embodiment is an oil and gas installation with a pipeline system including the control system described above.

According to a second aspect of the present invention there is provided a method of control of a dynamic system by means of a control loop incorporating a baseline controller and an autonomous supervisory layer, the baseline controller including an adaptive control layer, the method comprising: using the baseline controller to control a manipulated variable to adjust a controlled variable in accordance with a setpoint for the controlled variable: using the adaptive control layer for providing adaptive feedback control, the adaptive control layer being arranged to adjust the control of the manipulated variable in response to changes in the dynamics of the controlled system by adjusting parameters of the baseline controller and/or by adjusting the control input for the manipulated variable; and using the supervisory layer for monitoring performance and stability of the control loop and for adjusting the setpoint, wherein the autonomous supervisory layer moves the setpoint toward a more stable value when potential instability is identified and moves the setpoint toward a potentially less stable but higher performing value when the control loop is stable, wherein the supervisory layer checks for oscillations in the system and, if there are no oscillations within set thresholds, the setpoint is adjusted toward a higher performing value, wherein the supervisory layer is arranged to identify variations in parameter(s) of the system that indicate a potential problem and to adjust the setpoint toward a more stable value in response to potential problems, and wherein the changes to the setpoint by the supervisory layer provide excitations to the dynamic system to enable the adaptive controller to work effectively.

As with the system described above, the controlled variable may be a valve opening and the manipulated variable may be a pressure upstream of the valve, with the valve being at the top of a riser in a pipeline system for multiphase fluid flow.

Thus, a preferred example is a method of reducing slug flow in a pipeline for multiphase fluid, where the pipeline includes an inclined section and a valve that controls flow through the inclined section, the method comprising: using the baseline controller to control an opening of the valve to adjust a pressure or other measurement within the pipeline valve in accordance with a setpoint; using the adaptive control layer to adjust the control of the valve opening in response to changes in flow conditions within the pipeline; and using the autonomous supervisory layer to move the setpoint to a more stable value when potential instability is identified and to move the setpoint to a potentially less stable but higher performing value when the control loop is stable. For example, the baseline controller may be used to control an opening of the valve to adjust a pressure within the pipeline upstream of the valve in accordance with a pressure setpoint or to adjust the valve opening in accordance with a valve opening setpoint and in this case the autonomous supervisory layer may be used to increase the pressure setpoint/decrease the valve opening setpoint when a potential instability is identified and to decrease the pressure setpoint/increase the valve opening setpoint when the control loop is stable.

The combination of adaptive feedback and an autonomous supervisory layer provides similar advantages to those discussed above for the control system of the first aspect. The method may include features equivalent to those discussed above for the control system.

The method may include using only a single measurement for the controlled variable, for example from a single sensor. This may be a pressure measurement as described above. The method may include the use of the same measured value by each of the baseline controller, the adaptive layer and the supervisory layer.

The method may include carrying out some or all of the following steps using the supervisory layer: detecting potential instability and to adjust the setpoint toward a more stable value, for example to a higher pressure/smaller valve opening; evaluating an integrated absolute error for the control loop over a time window and comparing it to a pre- specified threshold to assess the stability of the control loop; identifying growing oscillations with frequency in a certain predefined range, such as a frequency considerably lower than the average frequency of adjustment of the manipulated variable by the baseline controller; using a frequency analysis, for example a Fourier analysis, to identify oscillations at frequencies associated with potential problems, for example frequencies known to relate to slug flow when the control system is for a multiphase fluid flow system; detecting the impact of a major disturbance in the controlled system and to quickly adjust the setpoint to avoid problems such as too high a gain in the baseline controller, for example by checking for higher frequency oscillations, and or by identifying the frequency or frequencies with the highest amplitude(s) and compare against threshold values; carrying out a frequency spectrum analysis in order to identify the frequency or frequencies showing the highest amplitude, and checking the amplitude(s) against a threshold value as well as optionally checking the frequency(/ies) against a threshold.

In reaction to finding that there is a stable or unstable control loop then the method may include using the supervisory layer to adjust the setpoint as discussed above with reference to the control system of the first aspect. The threshold values in the steps set out above may be defined as discussed above in connection with the control system of the first aspect.

The method may include determining a control input to the manipulated variable as a summation of an output from the baseline controller with an adaptive augmentation control signal from the adaptive control layer. The baseline controller and adaptive control layer may be as discussed above.

The method may be a method for control of flow in an oil and gas installation in order to reduce slug flow.

Viewed from a further aspect the invention provides a computer programme product comprising instructions that, when executed, will configure a computer apparatus to perform the method of the second aspect. The computer programme product may incorporate the optional/preferred features of the second aspect as discussed above.

Preferred features of each aspect of the invention may be combined with the other aspects of the invention, and optionally with preferred features of the other aspects, as far as is applicable or appropriate.

Certain preferred embodiments of the invention will now be described by way of example only and with reference to the accompanying drawings in which:

Figure 1 shows a schematic representation of a pipeline and riser system;

Figure 2 illustrates an experimental setup used to test the slug reducing control system disclosed herein; Figure 3 is a simplified diagram of a proposed control system with an autonomous supervisory layer;

Figure 4 is a simplified block diagram of the proposed control system showing an adaptive control scheme;

Figure 5 shows experimental results from a first experiment using the proposed control system in conjunction with a well tuned loop transfer recovery (LTR) baseline controller;

Figure 6 shows parameters used in the adaptive controller during the experiment of Figure 5;

Figure 7 shows disturbances applied to airflow and water flow in a second experiment using the same control system;

Figure 8 shows the controlled and manipulated variables during the second experiment as the disturbances shown in Figure 7 are applied;

Figure 9 shows parameters used in the adaptive controller during the experiment of Figures 7 and 8;

Figure 10 shows the controlled and manipulated variables during a third experiment where the well tuned controller of the first two experiments is replaced by a poorly tuned proportional integral (PI) control and the adaptive controller is disabled;

Figure 1 1 shows the results of a fourth experiment with a similar baseline control to that of Figure 10, but with the adaptive control activated; and

Figure 12 shows a similar experiment to the third and fourth experiments, with the adaptive control activated and the poorly tuned baseline controller replaced with the well tuned LTR controller of the first experiment.

As a way to demonstrate the proposed control system the discussion herein makes reference to a simplified pipeline and riser system as shown schematically in Figure 1 , and as shown in experimental form in Figure 2. It will be appreciated, however, that the same control system could be applied to other systems, for example other two-phase flow systems such as two-phase refrigeration circuits and the like. Also, the same control system has advantages for a pipeline with any inclined section, such as a production pipe, and could be applied to a valve at any location that controls flow through the inclined section, including at the bottom or at the top of the inclined section.

The basic parts of the example pipeline and riser system include a pipeline 12 feeding multiphase flow to a riser 14, with a choke valve 16 controlling pressure at the top of the riser 14 and feeding fluid into an outlet pipe 18. The experimental setup shown in Figure 2 includes the same basic parts with the addition of a separator 20 after the outlet pipe 18, and a multiphase flow generator 22 arranged to feed multiphase flow into the pipeline 12. For the purposes of the experiments the multiphase flow was water and air, but of course it will be appreciated that the same conclusions apply to multiphase hydrocarbons in an oil and gas production installation. It should be noted that the separator 20 in the experimental setup is at a similar position to a typical buffering separator in a real world oil and gas installation.

Anti-slug control in multiphase risers involves stabilizing an open-loop unstable operating point. Existing anti-slug control systems are not robust and tend to become unstable after some time, because of inflow disturbances or plant dynamic changes, thus, requiring constant supervision and retuning. A second problem is the fact that the ideal setpoint is unknown and we could easily choose a suboptimal or infeasible operating point. In this paper we present a method to tackle these problems. The control solution described herein is composed of an autonomous supervisor that seeks to maximize production by manipulating a pressure setpoint and a robust adaptive controller that is able to quickly identify and adapt to changes in the plant. The supervisor is able to automatically detect instability problems in the control loop and moves the system to a safer, stable operating point. This solution has been tested with the experimental rig of Figure 2 and the results show significant improvements.

The severe-slugging flow regime which is common at offshore oilfields is characterized by large oscillatory variations in pressure and flow rates. This multi-phase flow regime in pipelines and risers is undesirable and an effective solution is needed to suppress it . One way to prevent this behaviour is to reduce the opening of the top-side choke valve. However, this conventional solution reduces the production rate from the oil wells. A known recommended solution to maintain a non-oscillatory flow regime together with the maximum possible production rate is active control of the topside choke valve. Measurements such as pressure, flow rate or fluid density are used as the controlled variables and the topside choke valve is the main manipulated variable.

From an economic point of view, one would like to have the lowest possible pressure (maximum valve opening) in the pipeline/riser system. This translates into low pressures at the bottom hole of the wells which maximizes the fluid inflow from the reservoir. However, as the pressure setpoint decreases the stabilization of the system becomes more difficult and, thus, the choice of the ideal setpoint is hard task. In fact, the ideal pressure setpoint is unknown and varies with the inflow conditions. Setting it too high reduces the production. Setting it too low may be infeasible (uncontrollable), leading to slug flow. Consequently, constant monitoring of the control system by the operators is needed.

Hence, it is proposed to use an autonomous supervisory system that safely drives the process in the direction of minimum pressure for production maximization. The main idea is to gradually decrease the pressure setpoint until just before the control performance is no longer acceptable due to slugging. The supervisor automatically assesses the performance and stability of the control loop and decides the direction in which we should change the pressure setpoint in order to ensure stable operation. For example, if slow oscillations with growing amplitude in the output are detected, the setpoint should be increased since it is safer and easier to stabilize.

Nonetheless, the standard linear controllers are typically designed for a given operating point and they may fail to give acceptable performance when the setpoint changes considerably. Another problem is disturbances in the inflow, which greatly affect the dynamics of the plant.

For these reasons the proposed solution further includes a robust adaptive anti-slug controller. One possible robust-adaptive feedback control design method is that proposed by Lavretsky (Lavretsky, E. (2012. Adaptive output feedback design using asymptotic properties of LQG/LTR controllers. IEEE Transactions on Automatic Control, 57(6), 1587- 1591 ). This method falls into the model-reference adaptive control category and fits well in the proposed approach. This controller is able to quickly identify and adapt to changes in the plant dynamics in order to recover the desired performance. Other similar control design methods could also be used.

The proposed control solution hence comprises both the autonomous supervisor and the robust adaptive slug control. The combination of these two elements results in a beneficial synergy: the periodic setpoint changes triggered by the supervisor introduce enough excitement in the system for the adaptation to work well; a well functioning adaptive controller allows the supervisor to push the system closer to the limit for a wide range of operating conditions.

It is worth to point out that this approach is very general and can be applied in a variety of applications with similar characteristics: dynamics change when approaching the (possibly unknown) operating limit of the system. Thus, although a pipeline-riser system as used in the oil and gas industry is one advantageous use for the proposed controller, it could also be used elsewhere, as discussed above.

Figure 1 shows a schematic representation for a pipeline-riser system. The inflow rates of gas and liquid to the system, w_{g in} and w _n, are assumed to be independent disturbances and the top-side choke valve opening (0 < Z < 100%) is the manipulated variable. A fourth-order dynamic model for this system was presented by Jahanshahi and Skogestad (Jahanshahi, E. and Skogestad, S. (201 1 ). Simplified dynamical models for control of severe slugging in multi-phase risers. In 18th IFAC World Congress, 1634-1639. Milan, Italy). The state variables of this model are defined as:

· m_gF: mass of gas in pipeline [kg]

m_lp: mass of liquid in pipeline [kg] m^: mass of gas in riser [kg]

m_lr mass of liquid in riser [kg]

The four state equations of the model are

r^{f = w}*™^~ ¾ (1 )

τ"_ίρ = w _n— w_t ^2)

" = ^W ₀ - ^aW (3)

m_i = w_¾— (1— )w (4)

The flow rates of gas and liquid from the pipeline to the riser, w_g and w are determined by pressure drop across the riser-base where they are described by virtual valve equations. The outlet mixture flow rate, , is determined by the opening percentage of the top-side choke valve, Z. The different flow rates and the gas mass fraction, a, in the equations (1])-(4) can be determined by suitable model equations, such as those given by Jahanshahi and Skogestad (ibid). In this paper we used the linearized version of that model for the control design methods. Alternatively, empirical low-order models could have been used.

An autonomous control system is proposed, with the purpose of driving the controlled process towards its operational limit. As set out above, the proposed solution is composed of two main elements:

a supervisory system that overlooks the control loop, assess stability and performance and makes a decision on which direction (increase or decrease) the setpoint should move. In the application, the strategy is to gradually reduce the pressure setpoint until a stability problem is detected (e.g., slow oscillations start to build-up). At this point the supervisor should move the system to a safer operating point (increase setpoint).

a robust adaptive controller that regulates the system to the setpoint specified by the supervisory controller. The controller must be able to identify changes in the plant dynamics and compensate for it to give acceptable closed-loop performance in a wide range of operating conditions.

Figure 3 shows a simplified representation of the supervisory control in conjunction with a robust adaptive controller, and Figure 4 shows more detail of the adaptive controller.

We believe that the combination of frequent setpoint changes by the supervisor with and adaptive control scheme can be very fruitful because the periodic setpoint changes triggered by the supervisor gives enough excitement in the system for the adaptation to work well; a well functioning adaptive controller allows the supervisor to push the system closer to the limit compared to linear controllers. A key component in an autonomous supervisor is the ability to quickly detect problems in the control loop. In the application the main problem is the appearance of slugging flow which is characterized by growing (slow) oscillations in the pressures and flows with a certain frequency. Such oscillations are a signal that the controller is having problems to control the process at the given operating conditions and should move to a safer setpoint. Algorithm 1 (described below) exemplifies a basic supervisory scheme for the anti-slug control problem. is the pressure setpoint and AP^ represents the size of the steps. The pressure can be measured at any point of the system (e.g. riser base or riser top). Note that the amplitude of the step when increasing or decreasing the setpoint may be different.

The basic idea is to periodically check for slow oscillations in the system and decrease the setpoint only if nothing is detected. On the other hand, one should quickly increase the setpoint if the amplitude of the oscillations is starting to grow. In this case, it could be desirable to reset the adaptation parameters to the previous good values using, for instance, a look-up table.

For a practical application, however, many other safeguards must be included. For example, if a major disturbance occurs, the controlled variable may drift away from the setpoint very rapidly and the oscillation detection system may fail to perceive in time. In order to quickly detect these major problems a second, independent check function must be implemented. In this example the mean control error is periodically analysed over a short time horizon. A warning flag is raised if the mean error is increasing too quickly or if it crosses some large threshold. The system should also include a routine to detect high frequency oscillations generally caused by having too high control gains for the given operating conditions. In this case one should decrease the setpoint instead. Other functions of the supervisor could include looking after the adaptive control (e.g. we may want to turn off the adaptation during the starting up period), fault detection, alarms, etc.

Algorithm 1

Loop

Analyse measured data

if slow oscillation detected then

if amplitude is increasing then

Return to previous adaptation values

else wait longer

end if

else end if

end loop

A key component in an autonomous supervisor is the ability to quickly detect slow oscillations in the closed-loop system. This can be achieved by periodically applying a frequency analysis tool in the measured data (e.g. pressures) in a moving-horizon manner. The chosen approach is to estimate the power spectral density using a fast Fourier transform and then check if the main frequency component of the signal lies in a neighbourhood of the slug frequency. If this is the case, a warning flag is raised. The same frequency analysis can be used to estimate the amplitude of the oscillation, allowing us to tell whether the oscillations are increasing or fading out.

The practical experience has shown that this approach is quite robust and it only requires knowledge of the slug frequency for the specific application. No other tuning parameters are necessary.

The proposed system uses a robust adaptive output feedback design method as proposed by Lavretsky (Lavretsky, E. (2012. Adaptive output feedback design using asymptotic properties of LQG/LT controllers. IEEE Transactions on Automatic Control, 57(6), 1587-1591 ). This method falls into the model-reference adaptive control category . The main components of this controller are: an observer-like reference model which specifies the desired closed-loop response; a linear baseline controller that gives the desired performance and robustness at nominal conditions; the adaptation law which augments the input in order to recover the desired performance despite the disturbances and uncertainties (See Figure 4). For completeness, we will outline in the following the design method that was used. This follows the notation used in Lavretsky, E. and Wise, K. (2013), Robust and Adaptive Control with Aerospace Applications, published by Springer.

We assume that system can be described in the following form

x = Ax + BA(u + θ^Γ ζπ}) + ¾_{Zsp (5)} where A E R^nXTi , B E R^nXTn, C E R**" and C_z E R^mXm are known matrices. Note that the matrices may have been augmented to include the integral feedback connections. The vector x E R represents the system states, y E R^p are the available measurements, u ε R^m are the inputs and∑ e R^m are the variables we wish to regulate to given setpoints z^. The uncertainties are described by an unknown diagonal matrix A, an unknown matrix of coefficients B and a known Lipschitz-continuous regressor Φ(^χ). We assume that the number of available measurements p is larger than the number of control inputs m. In this case, the system can be 'squared-up' using pseudo-control signals to yield minimum-phase plant dynamics.

Representation fits well with the application. One of the main challenges is the very large process gain variation as we change the pressure setpoint. This can be represented by A. Furthermore, the poles and zeros of the linearized dynamics move considerably as the pressure reduces. This effect can be modelled by the term © ΤΨ (x) as long as we make a good choice for the regressor <£(.x).

The first step is to design a reference model with the desired closed-loop dynamics. In this case we compute an optimal state feedback K_LiQS by employing the LQR method such that

^Arrf ^{= A} — ^BKLQR (6)

as the desired dynamic characteristics. It has been shown that the transient dynamics of the adaptation scheme can be improved by using an observer-like model reference. Thus, the reference model becomes

x where L_v E R^nXm j_s the prediction error feedback gain that is obtained by solving a certain algebraic Riccati equation (Lavretsky, 2012, ibid). The 'square-up' step of the plant dynamics should be performed prior to the design of L_v.

The chosen implementation approach is to augment a baseline linear controller with the adaptor instead of using a fully adaptive control. The reasoning comes from the fact that in most realistic applications a stabilizing baseline controller might already be in place. This baseline controller would have been designed to give satisfactory performance under nominal conditions around an operating point. If the performance degrades due to changes of operating conditions, we will attempt to recover the desired performance by augmenting the baseline controller with an adaptive element. The total control input is the sum of the components

« = «»! + ¾d (8)

where m_hl denotes the baseline control input and _ad is the adaptive augmentation control signal.

The adaptation increment n_ad is given by

where is an estimation of © and Jf_u serves as an estimate of (l_{m Xm}— Λ^{" 1}). Given the adaptation rates Γ_θ and F_m the adaptive law with the Projector

Modification (Pomet, J.B. and Praly, L. (1992). Adaptive nonlinear regulation: estimation from the lyapunov equation. I EEE Transactions on Automatic Control, 37(6), 729) can be written as

Pr₀j{§ ~T_e #i_:¾re ¾-^ft5 WS ^T = Proj(R_u, -Γ_Αί¾¾ο WS^T)

dt ^" - ~ ^' * - ^' (1 1 ) where

ey ^{= y}"f ^{~ y} (12) is the output tracking error and the matrices J?₀, W and S are selected to ensure that the tracking error e_y becomes small in finite time.

The projector operator Pr oj ensures that the adaptive parameters always lie inside a user-defined region and can never diverge. The robustness of this adaptive law can be improved by including a dead-zone modification that stops adaptation when the error e_y is too small. Such modification ensures that the adaptation parameters will not drift because of measurement noise, as explained further by Lavretsky and Wise (2013, ibid).

It is interesting to note that upon combining and we get

u = (1 - Jf_H}i¾_. - §-^T(fi(x_rgf) _{(1 3)} where we see that the adaptor is in essence modifying the baseline controller gain by a factor ( 1— !t_u). The second term in the right-hand side of the equation tries to match and cancel the effect of the nonlinear uncertainties in equation (5).

The observer-based model reference works as a robust closed-loop Luenberger estimator when we select the baseline controller:

This leads to an output feedback controller equivalent to the loop transfer recovery using the Lavretsky method, which has been proven to have excellent robustness properties (Lavretsky and Wise, 2013, ibid). This is the baseline controller of choice because of its robustness properties and its good performance observed in the experiments. Nonetheless, any other linear controller (e.g. PI control) could have been selected for the baseline layer. In fact, experiments have shown that the adaptive control scheme presented above is able to recover the desired performance even if a poorly tuned PI controller is used in the baseline (See Figures 10 and 1 1 ). Another advantage of using the augmentation approach for the adaptive scheme (rather than fully adaptive control) is that the adaptation could be turned off when necessary without losing control of the system. This can be particularly important in some situations such as start-up.

Experiments were performed on a laboratory setup for anti-slug control at the

Chemical Engineering Department of NTNU. Figure 2 shows a schematic presentation of the laboratory setup. The pipeline and the riser are made from flexible pipes with 2 cm inner diameter. The length of the pipeline is 4 m, and it is inclined with a 15° angle at the bottom of the riser. The height of the riser is 3 m. A buffer tank is used to simulate the effect of a long pipe with the same volume, such that the total resulting length of pipe would be about 70 m.

The topside choke valve is used as the input for control. The separator pressure after the topside choke valve is nominally constant at atmospheric pressure. The nominal feed into the pipeline is assumed to be at flow rates 4 l/min of water and 4.5 l/min of air. With these boundary conditions, the critical valve opening where the system switches from stable (non-slug) to oscillatory (slug) flow is at Z ' = 15% for the top-side valve.

The desired steady-state (dashed middle line) in slugging conditions (Z > 15%) is unstable, but it can be stabilized by using control. The slope of the steady-state line (in the middle) is the static gain of the system, k = dy/ du = d P_in /dz. As the valve opening increase this slope decreases, and the gain finally approaches to zero. This makes control of the system with large valve openings very difficult.

The main parameter for the implementation of the supervisory controller is the period of the slug oscillation. This variable depends mainly on the dimensions of the pipeline and riser, although the operating conditions (e.g. valve opening) do have some effect on it. For the purposes it is enough to have an estimation of the order of magnitude of the frequency of the oscillations. In the application we observed variations in the oscillation period ranging from 40 to 70 seconds. Thus, any oscillation in this frequency range will be reported by the oscillation detection algorithm. The core idea of the supervisor is Algorithm 1. The loop was executed every 20 seconds to avoid strong interactions with the stabilizing control layer. The length of the horizon for analyses in the oscillation detector was set to 90 seconds to ensure that a full slug cycle would be detected.

We designed the controllers based on the linearized version of the model described above for a valve opening Z— 30%. In the control algorithm we consider measurements of both the inlet pressure of the pipeline (P_iR) and the pressure in the riser top (P_rC). The regulated output in experiments is z = P_i . The second measurement is used to ensure robustness properties of the LTR baseline and the adaptive controllers. In the application we chose the (14) as the baseline controller because of its excellent robustness properties and its good performance observed in the experiments. Prior to conducting the LQG/LTR controller design, we augmented the plant dynamics to include the integrated inlet pressure tracking error e = P_sp— P_i7l.

For the adaptive algorithm we chose as basis function the linear relationship where P_i7l is an estimation of the inlet pressure. From the analysis this simple basis function is enough to describe the variation in the plant dynamics (zeros and poles) due to changes in the operating point (indicated by P_in). The gain uncertainty is described by the unknown scalar parameter A. Therefore, the adaptation scheme is composed of two scalar adaptive parameters only. The Projector Operator ensures that these parameters are bounded and remain inside the interval [—5, 5].

To improve the quality of the adaptation and to ensure the overall robustness of the system, we switched on the adaptation only after a setpoint change is made and for a limited amount of time (e.g. for 1 min). This prevents the system to wrongly adapt to the disturbances. When the supervisory layer detects a problem in the system and the setpoint is increased, the adaptation parameters are reset to the closest previously computed value for the given setpoint using a lookup table.

For comparison we have also implemented a PI controller in the baseline layer. The experiments have shown that the adaptive control scheme presented in the previous section is able to recover the desired performance even if a poorly tuned PI controller is used in the baseline (See Figures 10 and 1 1 ).

I n a first experiment with nominal flow conditions the feed into the pipeline is set to be at constant flow rates, 4 l/min of water and 4.5 l/min of air. Figure 5 depicts the results for a 48 minutes run of the complete system. The setpoint is indicated by the dashed line in the top plot. Note that the setpoint is only decreased when the supervisor is sure it is safe. The detection of growing oscillations is indicated by the warning flag. In Figure 5 these can be seen around the times 15.5, 27, 34 and 42 minutes. The supervisor is able to safely keep the system at stable conditions at fairly high valve openings. Figure 7 shows the adaptation parameter for the same experiment. The adaptation is switched on after 100 seconds to avoid the start-up dynamics. It is interesting to note that at first the parameter &_u increases (the gain (1— if_u) decreases) indicating that initially the controller is a bit too aggressive for the given conditions. However, as the supervisor reduces the setpoint for P_f„ the parameter &_v decreases (the gain (1— itj increases) considerably to maintain the desired performance. Note that we reset the adaptation parameters when a problem is detected (warning flag). In a second experiment we tested the more realistic and challenging conditions in which the gas to liquid ratio varies considerably throughout the experiment. Initially the feed into the pipeline is set to constant flow rates 4 l/min of water and 4.5 l/min of air. Then, a sequence of steps in the air flow is applied: first we increase the air flow by 50% at t = S min followed by a 30% decrease at t = 20 min (see Figure 7). Changes in the air flow and pressures naturally perturb the water flow. Note that these changes represent very serious disturbances that have big effect in the dynamics of the plant, and thus they simulate the effect of significant flow problems in a real world system.

Figure 8 shows the performance of the control system. The more serious disturbance here is when the air flow decreases ( t = 20 min). The pressure rapidly diverges since it became very difficult to stabilize the system at these conditions.

Nonetheless, the supervisory layer quickly detected the problem and immediately moved the system to a safer operating point. After stabilizing the process, the robust adaptive controller was able to adapt its parameters for the new dynamics (see Figure 9), making it possible to reduce the pressure setpoint even under such harsh conditions. It is worth to point out that slugging flow did not occur at any moment and the good performance of the controller remained consistent, proving the great resilience of the proposed solution. Such a result would not have been possible to achieve without an autonomous supervisor and an adaptive controller.

For comparison, it is interesting to investigate the effect of the baseline controller in the overall performance of the control system. The incentive for doing so is clear: in most realistic applications a stabilizing baseline controller might already be in place and perhaps we do not want to change it.

For this purpose we consider as the baseline a poorly tuned PI controller in a third experiment. Figure 10 shows the results of the autonomous supervisor with the PI controller without any adaptation. We observe an overall poor performance and the inability to operate with large valve openings.

The experiment was repeated in a fourth experiment with the same PI controller but with the adaptation switched on. The same reference model used in the first and third experiments discussed above is employed here. Figure 1 1 shows the results. The closed- loop performance was greatly improved compared to Figure 10 and we are able to operate at a larger valve opening. For a complete comparison, we ran a fifth experiment similar to the fourth experiment but experiment using the well tuned LTR controller with adaptation switched on. Figure 12 shows the result of this controller where we observe good tracking performance throughout the experiment. Table 1 summarizes the results of third, fourth and fifth experiments where we compare the tracking performance based on the integrated square error (ISE) and the 'economic' performance based on the mean valve opening and pressure. Note that the improvement from experiment 3 to 4 is substantial, where we observe an increase of 31 % of the average valve opening. On the other hand, the improvement from experiment 4 to 5 is only minor. Nevertheless, the recommendation is to always use a good robust controller in the baseline. This will ensure safer operation during start-up (when the adaptation is likely to be turned off) or during reset of the control system.

The adaptive controller described in the above examples makes use if the measurement of both top and bottom riser pressure. It will be appreciated that a similar controller could be provided that required only one of these pressure measurements.

An autonomous control system is described herein that can improve performance of two phase flow systems and, for example, can maximize oil production in off-shore oilfields. The proposed control solution includes an autonomous supervisor that manipulates the pressure setpoint and a robust adaptive controller that is able to quickly identify and adapt to changes in the plant. The supervisor has been shown to be able to automatically detect instability problems in the control loop and move the system to a safer operating point when necessary. The experimental results show good performance and great resilience in a variety of operating conditions. The proposed solution will lessen the demand for manual supervision, will reduce the need for frequent retuning of the controller and will maximize the oil production, as well as offering improvements in other similar two-phase flow situations.

Claims

1. A control system for controlling a dynamic system, the control system comprising: a control loop incorporating a baseline controller with an adaptive control layer, wherein the baseline controller is for controlling a manipulated variable to adjust a controlled variable in accordance with a setpoint for the controlled variable, and wherein the adaptive control layer is for providing adaptive feedback control, the adaptive control layer being arranged to adjust the control of the manipulated variable in response to changes in the dynamics of the controlled system by adjusting parameters of the baseline controller and/or by adjusting the control input for the manipulated variable; and

an autonomous supervisory layer for monitoring performance and stability of the control loop and for adjusting the setpoint, wherein the autonomous supervisory layer is arranged to move the setpoint toward a more stable value when potential instability is identified and to move the setpoint toward a potentially less stable but higher performing value when the control loop is stable, wherein the supervisory layer is arranged to check for oscillations in the system and, if there are no oscillations within set thresholds, to adjust the setpoint toward a higher performing value, wherein the supervisory layer is arranged to identify variations in parameter(s) of the system that indicate a potential problem and to adjust the setpoint toward a more stable value in response to potential problems, and wherein the changes to the setpoint by the supervisory layer provide excitations to the dynamic system to enable the adaptive controller to work effectively.

2. A control system as claimed in claim 1 , wherein the manipulated variable is a valve opening and the controlled variable is a pressure upstream of the valve, with the valve being at the top of a riser in a pipeline system for multiphase fluid flow.

3. A control system as claimed in claim 1 , wherein the control system is for reducing slug flow in a pipeline for multiphase fluid, where the pipeline includes an inclined section and a valve that controls flow through the inclined section, the control system comprising: the control loop incorporating the baseline controller with the adaptive control layer, wherein the baseline controller is for controlling an opening of the valve to adjust a pressure or other measurement within the pipeline in accordance with a setpoint, and wherein the adaptive control layer is for providing adaptive feedback control, the adaptive control layer being arranged to adjust the control of the valve opening in response to changes in flow conditions within the pipeline; and

the autonomous supervisory layer for monitoring performance and stability of the control loop and for adjusting the setpoint, wherein the autonomous supervisory layer is arranged to move the setpoint to a more stable value when potential instability is identified and to move the setpoint to a potentially less stable but higher performing value when the control loop is stable.

4. A control system as claimed in claim 1 , 2 or 3, wherein the baseline controller, the adaptive layer and the supervisory layer each make use of the same measured value(s) relating to the controlled variable.

5. A control system as claimed in claim 1 , wherein the supervisory layer is arranged to identify a potential problem by evaluating the integrated absolute error for the control loop over a time window and compare it to a pre-specified threshold for integrated absolute error to assess the stability of the control loop.

6. A control system as claimed in claim 5, wherein the pre-specified threshold for integrated absolute error is a value set as to provide a clear distinction between a potential problem and normal/expected noise in the system, for example a value that is at least two, three or four times the integrated absolute error values that arise with noise.

7. A control system as claimed in any preceding claim, wherein the supervisory layer is arranged to identify a potential problem by identifying growing oscillations with a frequency in a certain predefined range, typically a frequency considerably lower than the average frequency of adjustment of the manipulated variable by the baseline controller.

8. A control system as claimed in claim 7, wherein the frequency in a certain predefined range is set based on a known, predicted, or modelled slug frequency for the system and, for example, is set as a multiple of that frequency.

9. A control system as claimed in any preceding claim, wherein the supervisory layer is arranged to identify a potential problem by using a frequency analysis to identify oscillations at frequencies associated with potential problems, for example frequencies known to relate to slug flow when the control system is for a multiphase fluid flow system.

10. A control system as claimed in any preceding claim, wherein when the supervisory layer detects a potential problem indicating a possible instability then the setpoint is adjusted toward a more stable value, for example by changing the setpoint to adjust a valve opening by at least or about 2%, 3%, 5% or 10%.

1 1. A control system as claimed in any preceding claim, wherein the supervisory layer is arranged also to detect the impact of a major disturbance in the controlled system and to adjust the setpoint to avoid high amplitude frequency oscillations caused by too high a control gain and/or to avoid quickly developing instabilities when the pressure diverges swiftly from the setpoint, which may occur after a large change in operating conditions.

12. A control system as claimed in claim 1 1 , wherein the supervisory layer is arranged to check for higher frequency oscillations, and/or to identify the frequency or frequencies with the highest amplitude(s) and compare against threshold values to thereby identify possible instances of too high a control gain, and to shift the set point to a more stable value, for example by changing the setpoint to adjust a valve opening by at least or about 2%, 3%, 5% or 10%.

13. A control system as claimed in claim 12, wherein the supervisory layer is arranged to carry out a frequency spectrum analysis in order to identify the frequency or frequencies showing the highest amplitude, and then check the amplitude(s) against a threshold value as well as optionally checking the frequency(/ies) against a threshold frequency.

14. A control system as claimed in claim 13, wherein the threshold amplitude is set based on a multiple of an amplitude of oscillations due to noise, which could be based on a measured, predicted or expected value for the amplitude of oscillations due to noise.

15. A control system as claimed in claim 13 or 14, wherein the supervisory layer is arranged to identify possible instances of too high a control gain when there is an oscillation at a frequency over the threshold frequency and with an amplitude above the threshold amplitude value.

16. A control system as claimed in claim 13, 14 or 15, wherein the system is used to reduce slug formation in a pipeline with multiphase flow and the threshold frequency is defined with reference to a frequency of slug formation for the system without active control of the valve opening.

17. A control system as claimed in any of claims 11 to 16, wherein in order to identify quickly developing instabilities the supervisory layer is arranged to check the rate of change of the integrated absolute error for the control loop over a time window, and to identify a possible quickly developing instability when the integrated absolute error is increasing a pre-specified threshold for integrated absolute error despite corrections and/or when the rate of change of the integrated absolute error is above a threshold rate.

18. A control system as claimed in claim 17, wherein in reaction to potential quickly developing instabilities the supervisory layer is arranged to adjust the setpoint to a more stable value and the adjustment is larger than the adjustment claimed in claim 12 or claim 14, for example by changing the setpoint to adjust a valve opening by at least or about 10%, 20% or 30%.

19. A control system as claimed in any preceding claim, wherein when adjusting the setpoint to a more stable value the supervisory layer also directs an adjustment to the adaptation values set by the adaptive control layer, for example a return to the values being used prior to the start of the instability.

20. A control system as claimed in any preceding claim, wherein a pre-existing baseline controller is augmented with an adaptive layer.

21. A control system as claimed in any preceding claim, wherein the control input to the manipulated variable is a summation of the output from the baseline controller and an adaptive augmentation control signal from the adaptive control layer.

22. A control system as claimed in any preceding claim, wherein the adaptive control layer is based on a reference model for the controlled system, the reference model being a model with similar dynamic characteristics to the controlled system.

23. A control system as claimed in any preceding claim, wherein the controlled system is a pipeline with an inclined section and the manipulated variable is the opening of a choke valve of the pipeline system.

24. A control system as claimed in claim 23, wherein the pipeline system is a part of an oil and gas installation.

25. An oil and gas installation with a pipeline system including the control system of any preceding claim.

26. A method of control of a dynamic system by means of a control loop incorporating a baseline controller and an autonomous supervisory layer, the baseline controller including an adaptive control layer, the method comprising:

using the baseline controller to control a manipulated variable to adjust a controlled variable in accordance with a setpoint for the controlled variable:

using the adaptive control layer for providing adaptive feedback control, the adaptive control layer being arranged to adjust the control of the manipulated variable in response to changes in the dynamics of the controlled system by adjusting parameters of the baseline controller and/or by adjusting the control input for the manipulated variable; and

using the supervisory layer for monitoring performance and stability of the control loop and for adjusting the setpoint, wherein the autonomous supervisory layer moves the setpoint toward a more stable value when potential instability is identified and moves the setpoint toward a potentially less stable but higher performing value when the control loop is stable, wherein the supervisory layer checks for oscillations in the system and, if there are no oscillations within set thresholds, the setpoint is adjusted toward a higher performing value, wherein the supervisory layer is arranged to identify variations in parameter(s) of the system that indicate a potential problem and to adjust the setpoint toward a more stable value in response to potential problems, and wherein the changes to the setpoint by the supervisory layer provide excitations to the dynamic system to enable the adaptive controller to work effectively.

27. A method as claimed in claim 26, wherein the manipulated variable is a valve opening and the controlled variable is a pressure upstream of the valve, with the valve being at the top of a riser in a pipeline system for multiphase fluid flow.

28. A method of reducing slug flow in a pipeline for multiphase fluid comprising a method as claimed in claim 26, where the dynamic system is a pipeline including an inclined section and a valve that controls flow through the inclined section, the method comprising:

using the baseline controller to control an opening of the valve to adjust a pressure within the pipeline upstream of the valve in accordance with a pressure setpoint or to adjust the valve opening in accordance with a valve opening setpoint;

using the adaptive control layer to adjust the control of the valve opening in response to changes in flow conditions within the pipeline; and

using the autonomous supervisory layer to move the setpoint to a more stable value when potential instability is identified and to move the setpoint to a potentially less stable but higher performing value when the control loop is stable.

29. A method as claimed in claim 26, 27 or 28 comprising use of a control system as claimed in any of claims 1 to 24.

30. A method as claimed in any of claims 26 to 29, comprising use of the same measured value by each of the baseline controller, the adaptive layer and the supervisory layer.

31. A method as claimed in any of claims 26 to 30, comprising carrying out some or all of the following steps using the supervisory layer: detecting a potential instability and adjusting the setpoint toward a more stable value, for example to a higher pressure/smaller valve opening; evaluating an integrated absolute error for the control loop over a time window and comparing it to a pre-specified threshold to assess the stability of the control loop; identifying growing oscillations with frequency in a certain predefined range, such as a frequency considerably lower than the average frequency of adjustment of the manipulated variable by the baseline controller; using a frequency analysis, for example a Fourier analysis, to identify oscillations at frequencies associated with potential problems, for example frequencies known to relate to slug flow when the control system is for a multiphase fluid flow system; detecting the impact of a major disturbance in the controlled system and quickly adjusting the setpoint to avoid problems such as too high a gain in the baseline controller, for example by checking for higher frequency oscillations, and or by identifying the frequency or frequencies with the highest amplitude(s) and comparing against threshold values; carrying out a frequency spectrum analysis in order to identify the frequency or frequencies showing the highest amplitude, and checking the amplitude(s) against a threshold value as well as optionally checking the frequency(/ies) against a threshold.

32. A computer programme product comprising instructions that, when executed, will configure a computer apparatus to control a control loop incorporating a baseline controller with an adaptive control layer and an autonomous supervisory layer to perform the method of any of claims 26 to 31.