(19)
(11)EP 2 734 900 B1

(12)EUROPEAN PATENT SPECIFICATION

(45)Mention of the grant of the patent:
16.09.2020 Bulletin 2020/38

(21)Application number: 12742838.1

(22)Date of filing:  20.07.2012
(51)International Patent Classification (IPC): 
G05B 13/04(2006.01)
(86)International application number:
PCT/EP2012/064323
(87)International publication number:
WO 2013/011134 (24.01.2013 Gazette  2013/04)

(54)

REGULATION METHOD

REGULIERUNGSVERFAHREN

PROCÉDÉ DE RÉGULATION


(84)Designated Contracting States:
AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

(30)Priority: 20.07.2011 EP 11290333

(43)Date of publication of application:
28.05.2014 Bulletin 2014/22

(73)Proprietor: General Electric Technology GmbH
5400 Baden (CH)

(72)Inventors:
  • HISSEL, Anne-Marie
    F-90018 Belfort (FR)
  • DE LARMINAT, Phillipe
    F-44000 Nantes (FR)

(74)Representative: BRP Renaud & Partner mbB Rechtsanwälte Patentanwälte Steuerberater 
Königstraße 28
70173 Stuttgart
70173 Stuttgart (DE)


(56)References cited: : 
EP-A2- 0 389 281
US-A1- 2004 167 657
US-A- 5 374 011
US-A1- 2006 075 981
  
      
    Note: Within nine months from the publication of the mention of the grant of the European patent, any person may give notice to the European Patent Office of opposition to the European patent granted. Notice of opposition shall be filed in a written reasoned statement. It shall not be deemed to have been filed until the opposition fee has been paid. (Art. 99(1) European Patent Convention).


    Description


    [0001] The invention concerns a method for regulation of an automatic system, applicable in particular to a device for regulation of the stator voltage of an alternating current generator.

    [0002] Alternating current generators, in particular of high power (several hundred megawatts (MW)), are connected to electricity distribution networks the demand of which varies greatly.

    [0003] These generators are subjected to varied disturbances of very different kind and magnitude: short circuits, voltage drops, load variation, load shedding, etc. In all cases, and throughout their operating range, performance as close as possible to the optimum is expected. Closed loops must also have sufficient stability margins.

    [0004] The regulation methods used at present, in particular for high-power alternators in nuclear power stations, are based on the so-called four-loop regulator principle, in which feedback (FBK) loops are used to maintain the output values as close as possible to a reference value, notably by controlling a certain number of controllable parameters.

    [0005] US 2006/075981 A1 discloses a control duty which asymptotically makes the output of a control subject consistent with an output of a reference model. The output of the control subject simulates a dynamic characteristic of a variable valve timing controller, and the output of the reference model simulates an ideal input-output characteristic of the variable valve timing. The control duty makes a difference between the output of the reference model and an actual valve timing small enough.

    [0006] These methods based on analogue technologies are highly sensitive to measurement errors and are even relatively ineffective in assuring the stability of closed loops over a wide range. In particular, these closed loop methods generate oscillations that are difficult to damp out and often poorly damped.

    [0007] These regulation methods and the regulators applying them more particularly fail to meet the technical specifications of electricity suppliers relating to exciter and voltage adjustment equipment of high-power alternators in nuclear power stations over the whole of the range of use.

    [0008] In order at least partially to alleviate the shortcomings previously referred to, the invention consists in a method for automatic regulation of a system in which a plurality of parameters characteristic of the system are measured and in which at least one control parameter is applied as a function of the measured parameters,
    characterized in that:
    • a nominal operating point of the system is chosen,
    • a nominal model describing the system at this nominal operating point is determined,
    • a set of models representative of possible variations relative to the nominal model is determined,
    • estimated characteristic output parameters corresponding to measurable characteristic parameters are determined from said nominal model,
    • the errors between at least one of the measured characteristic output parameters and at least one of the estimated characteristic output parameters are determined, and
    • at least one control parameter of the system is applied as a function of the errors determined to reduce the error between at least one of the estimated characteristic output parameters and the corresponding at least one measured value of the measured characteristic output parameters.


    [0009] The resulting method does not use an estimator to obtain its reference values, which renders it less sensitive to parametric uncertainties in respect of the non-measurable magnitudes relating notably to the status of the system.

    [0010] The method may also have one or more of the following features, separately or in combination.

    [0011] The nominal model is an approximation of the ideal transfer function of the system.

    [0012] The nominal model is the linearization of the ideal transfer function of the system around an operating point.

    [0013] Said at least one control parameter of the system applied as a function of the errors determined to reduce the error between the estimated characteristic output parameters and the measured characteristic output parameters is determined by applying optimum feedback to an augmented system derived from the initial system by adding the integral to at least one measurable predetermined output parameter.

    [0014] The control parameter of the system that is applied is determined by optimization of an integral criterion.

    [0015] The control parameter of the system that is applied is determined by the Linear Quadratic Gaussian (LQG) optimum control method.

    [0016] The method further includes the following steps:
    • the error of the nominal model of the system is parameterized by decomposition over all the errors between the models of the set of models representative of the possible variations and the nominal model,
    • a given optimization criterion is minimized by varying at least one of the previously obtained parameters of the error relative to the nominal model.


    [0017] The method further includes a supplementary step of optimization of the command with the error for the fixed nominal model by determination of at least one feedback gain.

    [0018] The steps of minimization of the optimization criterion by varying the parameters of the error relative to the nominal model of the system and optimization of the command with the error relative to the fixed nominal model by determination of at least one feedback gain are repeated successively in an iterative loop.

    [0019] The method includes a supplementary step of integration of the difference between the command and its saturated value, and the at least one control parameter of the system applied as a function of the errors determined to reduce the error between the estimated characteristic parameters and the measured values of the measurable characteristic output parameters is determined taking into account the integral of the difference between the control parameter and its saturated value.

    [0020] The system includes an electrical power station alternator connected to an electrical network and its exciter.

    [0021] The state of the system is represented by a state vector that includes the stator voltage, the rotation speed of the rotor, the total angle and an image of the flux in the exciter.

    [0022] The set of output magnitudes includes the stator voltage, the rotation speed of the rotor, the active power and an approximate value of the mechanical power modeling a main disturbance.

    [0023] Said at least one control parameter that is applied includes an approximate value of the mechanical power modeling the main disturbance and the exciter control voltage.

    [0024] Another object of the invention is a system for automatic regulation of a system in which a plurality of parameters characteristic of the system are measured and in which at least one control parameter is applied as a function of the measured parameters,
    characterized in that it includes means configured:
    • to choose a nominal operating point of the system,
    • to determine a nominal model describing the system at this nominal operating point,
    • to determine estimated characteristic output parameters corresponding to measurable characteristic output parameters from said nominal model,
    • to determine the errors between at least one of the measured characteristic output parameters and at least one of the estimated characteristic output parameters, and
    • to apply at least one control parameter of the system as a function of the errors determined to reduce the error between at least one of the estimated characteristic output parameters and the corresponding at least one measured value of the measured characteristic output parameters.


    [0025] Other features and advantages will become apparent on reading the description of the following figures, in which:
    • figure 1 is a diagram representing in flow chart form the steps of one embodiment of the method,
    • figure 2 is a block diagram representing one embodiment of the so-called nominal model of the system,
    • figure 3 is a block diagram representing one embodiment of the so-called "feedforward" predictive function of the invention,
    • figure 4 is a block diagram representing one embodiment of the so-called design model of the feedback system,
    • figure 5 is a block diagram representing the augmented system used for desensitization in one embodiment of the invention,
    • figure 6 is a block diagram representing a simplified second version of the augmented system from figure 5,
    • figure 7 is a functional block diagram representing an embodiment comprising the predictive function, feedback, desensitization and action on control saturation.


    [0026] The same references relate to the same elements in all the figures.

    [0027] The invention concerns a method for automatic regulation of a system. Figure 1 shows various steps of the method 100 for regulation of the system. The method is used in particular in the case of an alternator coupled to an electrical network. The objective is to apply the exciter voltage to the alternator in such a way as to assure the stability of the alternator whilst tracking a setpoint voltage. This setpoint voltage is established so as to track the demand of the network to which the alternator is connected.

    [0028] The alternator is of the turbo-alternator type, for example. It comprises a rotor driven by a turbine connected to the reactor and a stator. The stator is at a certain so-called stator voltage Vs.

    [0029] The first step 101 of the method 100 is the selection of a nominal model Mn, which can notably be the simplest, linear and invariant design model. This nominal model may in particular represent the linearization of the transfer function at a predetermined operating point, deemed to be that at which the system is deemed to be operating.

    [0030] The method is modeled around the operating point by the following set of equations:



    [0031] In the above equations:
    • x is a state vector, and in the case of the alternator x = [Vs ω θ efd]T where Vs is the stator voltage, ω the rotation speed, θ the total angle between the voltage of the network and the electromotive force, and efd is an image of the magnetic flux in the exciter,
    • um is an input vector, and in the case of the alternator um = [u Pmec]T where u is a control parameter and Pmec is a mechanical power communicated to the rotor, treated as the main disturbance,
    • y is the output vector, and in the case of the alternator y = [Vs ω Pe Pmec]T where Pe is the active electrical power supplied.


    [0032] Here the notation [...]T designates the transposition operation, the vectors being used in the form of vertical vectors in the formulas.

    [0033] A model is thus characterized by four matrices Am, Bm, Cm, Dm.

    [0034] The values of the various parameters are chosen to model the system optimally about a given operating point. That operating point is in general that around which the designer wishes to confer stability properties on the system to be regulated. For example, in the case of the alternating current generator this is the normal operating point.

    [0035] Am, Bm, Cm and Dm are matrices chosen to be invariant in the context of an invariant linear model. This invariant linear model produces a first approximation that is easy to manipulate and models the system around the operating point in a larger or smaller vicinity, depending on the required tolerance. This model is generally the first order linear approximation of the transfer function describing the real evolution of the system.

    [0036] It is possible to have time or other variable parameters contribute explicitly to the values of the matrices Am, Bm, Cm and Dm. In this case the calculations must take into account the values of the derivatives of these matrices. This quickly complicates the calculations, depending on the form of time dependency, but the method as a whole remains unchanged.

    [0037] Figure 2 represents in functional block diagram form the system modeled in this way.

    [0038] The central element of this functional block diagram 1 is the nominal model Mn, which comprises the alternator 3 and the exciter 5. The nominal model Mn receives as input the setpoint value Vref, the control parameter u and the main disturbance Pmec.

    [0039] As output, the nominal model Mn supplies the set of physical output magnitudes that includes the stator voltage Vs, from which the setpoint value Vref is subtracted in order to obtain the error e with respect to the setpoint, the electrical power Pe, and the vector y = [Vs ω Pe Pmec]T.

    [0040] The values of e and Pe are grouped in a vector z = [e Pe]T.

    [0041] The disturbances are for the most part of known kind and inherent to the physical implementation of the electrical power stations and networks, and in particular inherent to the fact that electrical power stations generally employ a certain number of alternators in parallel connected to a variable number of lines and consumers.

    [0042] A certain number of pertinent disturbances may be distinguished, including:
    • three-phase short-circuit: the voltage in the network falls sharply to zero over a short time period, with the result that the only reactance perceived is that of the transformer, after which the reactance of the line is re-established,
    • voltage dip: similar to a short-circuit, but for an intermediate network voltage drop value,
    • load shedding, the consequence of a prolonged short-circuit or voltage dip: the alternator is disconnected from all or part of the network; in the extreme case it no longer supplies power except to its auxiliaries to maintain its own operation,
    • loss of an adjacent set: in the context of a plurality of alternators in parallel, the failure or stopping of an adjacent alternator (set) can lead to under-exciter of the alternator concerned,
    • loss of at least one adjacent set at low voltage, leading to operation at the overexcitation limit: following the stopping of one or more adjacent sets, the alternator concerned switches to current limitation mode, and
    • frequency drop, caused by the loss of a mechanical power production site, which is reflected in a frequency drop of the order of a few hundred mHz in a period of a few seconds.


    [0043] The above disturbances are representative of those encountered in a real network, and it must be possible to eliminate them in the time scales set out in the technical specifications.

    [0044] In step 103 in figure 1, the nominal model is augmented by predictor models chosen to null for the setpoint and the main disturbance Pmec. The set of equations representing it is then as follows:

    with A11 = Am , A22 = 0, B1 = Bm1, Cy1 = Cm, De = Dy = 0

    and the vectors x1, = x , x2 = [Vref Pmec]T and ua = -Ga · x2



    [0045] Bm1 being the higher sub-matrix of Bm of appropriate size and Ga being a gain determined by solving the known Problem of Regulation with Internal Stability (PRIS), from which the form of the above equations stems. However, this gain may be obtained by other known regulator feedback methods.

    [0046] The augmented model is then used in the step 105 of figure 1 to reconstruct the state of the process. State reconstruction is usually based on an estimator such as a Kalman filter. Here, on the other hand, it is on the basis of the model used and the magnitudes measured that the state is reconstructed. The method thus uses the nominal model, here the design model, to establish estimated parameters that will serve as references. This function of the method is therefore referred to as feedforward (FFD) predictive action for the predictor aspect that it embodies through supplying reference magnitudes, as opposed to classic feedback.

    [0047] Thanks to FFD, state estimators are dispensed with. Moreover, having placed all the non-measurable magnitudes in said state vector, no further calculations are effected on them.

    [0048] In the step 107 of figure 1 optimum feedback is applied in order to deduce reference control parameters enabling optimum tracking of the setpoint to be obtained in the context of the nominal model. The linear feedback is effected notably by optimization of an integral criterion, which in the case of the alternating current generator may be:

    where Sr and Rr are positive weighting matrices.

    [0049] It is more particularly possible to apply a method such as the Linear Quadratic Gaussian (LQG) control method to effect this optimum feedback.

    [0050] Figure 3 shows in block diagram form the system with one embodiment of the FFD prediction loop.

    [0051] The central element of the figure 3 block diagram is the nominal model Mn.

    [0052] As input are received the mechanical power Pmec and the setpoint voltage Vref. The setpoint voltage Vref is filtered by a first order filter 7 with a known time constant Tref and thus with the transfer function (1 + sTref)-1.

    [0053] The reference command ur is determined from the filtered setpoint voltage, the mechanical power Pmec and the state x of the system. This reference command ur is supplied to the block Mn, which supplies as output the reference output vector yr. The reference or estimated output vector includes in the case of the alternator a reference stator voltage Vsr and a reference electrical power Per.

    [0054] On exit from the prediction loop there is obtained the set of estimated reference magnitudes, composed of the reference command ur and the reference output vector yr.

    [0055] It can be seen in said figure 3 that the optimum feedback used to obtain the reference command is of the form [G1; Ga+G1.Ta]. The gain G1 is obtained by optimization of the integral criterion on the basis of a command horizon Tr employing one of the usual methods. Ga and Ta are those obtained on classic solution of the PRIS problem, Ta being the integration horizon for the determination of Ga. The total gain of the optimum feedback used is the sum of these two terms.

    [0056] At least one of the parameters of the reference magnitudes ur and yr is then used in feedback (FBK) to determine an error relative to the nominal model. To this end, the errors between at least one of the measured characteristic output parameters y and at least one of the estimated characteristic output parameters yr are determined and at least one command parameter u of the system is applied or modified as a function of the errors determined to reduce the error between the estimated characteristic parameters yr and the measured values of the measurable characteristic parameters y.

    [0057] To effect this feedback FBK a so-called design model Mc shown in figure 4 is defined first.

    [0058] The central block of this diagram is the block combining the alternator 3 and the exciter 5, this time in their real form. This block receives as input the real command u and supplies as output the output magnitude vector y, from which the reference output vector yr is subtracted to obtain a vector ỹ = [Vs-Vsr Pe-Per] of the errors in the output relative to the reference. This vector ỹ is augmented to produce a vector Y by addition of the integral value of Vs-Vsr, by sampling Vs-Vsr and passage through an integrator 9.

    [0059] The standard model associated with the augmented process takes the form:

    in which E is a matrix enabling selection of the output parameter on which an integral action is to be introduced.

    [0060] Finally, the design model is chosen at an operating point of the process that can be the same as that for the FFD.

    [0061] The problem of optimization of the design system at the operating point is then solved.

    [0062] The solution may employ a known dual LQG/LTR control type regulator.

    [0063] The regulator then has two distinct functions: a function of reconstruction of the augmented state of the integral of the output voltage, and optimum linear feedback to the reconstructed augmented state.

    [0064] The invention makes provision for further improvement of the robustness of the command supplied by desensitization. To this end, the method is modeled around the nominal operating point chosen as follows:

    where w is Gaussian white noise retranscribing the state and measurement noise.

    [0065] The nominal model is completed in the step 109 in figure 1 with a set of K models {Mi} selected to be representative of the possible variations of the state of the system. Rigorously selected, these models form a variations "base".

    [0066] The error between any model from the set Mk and the nominal model MN is then parameterized by the projection of the difference M-MN onto the errors between the models of the set {Mi} and the nominal model MN:

    where δik varying from 0 to 1 is a normalized parameter.

    [0067] The production of the parameters δ can then be defined, as follows:



    [0068] In particular, the number of parameters δi is limited to the number m of underlying real parameters. The following set of equations can then be obtained to describe the evolution of the system:

    where ΦA1 = A1 - AN, ..., ΦB1 = B1 - BN, ..., Φc1 = C1 - CN, ..., ΦD1 = D1 - DN,

    and where



    [0069] The evolution of the system is then characterized by a set of equations represented in block diagram form in figure 5.

    [0070] In that figure, the central block 11 represents the standard system, characterized by Am, Bm, Cm, Dm, Q0, R0 and the set of models {Mk}.

    [0071] The uncertainties are transferred into an exterior loop of gain Δ.

    [0072] The optimum command at fixed Δ is determined by way of the loop of gain -K(s).

    [0073] Two systems are then defined. The first system H(s) encompasses the standard system 11 and the loop of gain -K(s). The second H(s) comprises the system H(s) and the loop of gain Δ.

    [0074] The system H(s) receives as input w and v, and supplies as output ζ and z. ζ is sent to the loop of gain Δ to obtain v (see above equations).

    [0075] Then, by defining Hζv, Hζw, Hzv, and Hzw, the dependent submatrices of K(s) of the transfer function H(s):



    [0076] This is translated on the complete system H(s) as:



    [0077] Assuming that ∥Δ∥< σ, where σ is an arbitrarily small adjustment parameter and ∥.∥ is a norm (the norm is 2 or infinite for example), the transfer function H may be developed as a Taylor series.

    [0078] Then H ≈ Hzw + HzvΔHζw + HzvΔHζvΔHζw

    [0079] The process to be optimized, represented in block diagram form in figure 6, is constructed by introducing the reconstructed vectors w̃ and ζ̃, in this particular case using the FFD predictor described above. It is also possible to use a reconstructed state obtained by another method, for example by means of a Kalman estimator.

    [0080] This augmented system receives as input w and w̃, combined in a vector W, separately on two parallel lines. w̃ is multiplied by σHζw. To obtain v. w and v are fed to the central block which represents the system 11. At the exit from the system-block 11 there are found ζ and z. ζ is multiplied by σHzv to obtain ζ̃, grouped with z in a single output vector Z.

    [0081] The augmented system also includes the feedback loop of gain -K(s) that connects the output y to the input u of the system-block 11.

    [0082] Starting with the non-desensitized regulator, in which the gain of the feedback loop has the value K0, the optimization of the system H then follows applied to Δ with fixed K(s) having the value K0. There is thus obtained a new system 11 to be optimized in terms of K(s) to determine a new gain K1 of this loop. With this new gain K1 a new system 11 is established by optimization applying to Δ. These latter steps are then repeated. δik are supplementary adjustment parameters. They can thus be only partly adjusted during the optimization steps. The choice of the δik that will be modified will essentially depend on the form of the models chosen.

    [0083] Hζw and Hzv depend on the value of K(s), and so on each iteration the dimension of K(s) increases. To prevent this it is possible, with a second approximation, to replace Hζw and Hzv with static gains g1, g2, by weighting ∥H∥ using σ.

    [0084] There is thus obtained an iterative loop which can be repeated until a convergence condition is satisfied. In practice, five repetitions have prove sufficient in most cases, and thus a fixed number of repetitions supplies an acceptable result.

    [0085] Figure 7 shows in functional block diagram form one embodiment of a regulator of an alternator 3, exciter 5 system as described above, further comprising a desaturation function.

    [0086] The real machine has only one bounded range of command u. The fact that the range is bounded is a result of the technical implementation of the system, and the value of the bound depends on the embodiment.

    [0087] In figure 7, the diagram comprises said system 11 with desensitization function, the predictor block FFD, the feedback function FBK, and a supplementary block 13 taking account of the saturation of the command u.

    [0088] The block FFD receives the main disturbance Pmec and the reference voltage Vref and supplies as output the reference magnitudes ur, yr.

    [0089] The block FBK receives as input the difference between the output of the system block 11 and the reference output yr, and with a feedback gain -K'(s) enables the command precursor ũ to be obtained, to which the reference command ur is added to obtain the command u, which after entering the supplementary saturation block 13 yields the saturated command usat that is supplied to the system 11.

    [0090] The command precursor ũ is obtained using a desaturation integrator 15 of gain 1/tau which integrates the difference between the command u and the saturated command usat.

    [0091] It should be noted in particular that the integrator 9 is placed in the block FBK, near its output, which corresponds to a change of variable relative to the FBK described above.

    [0092] The resulting regulator has demonstrated in simulations beneficial results concerning the removal of the disturbances referred to above in the context of simulations bearing in particular on the Flamanville EPR (European Pressurized water Reactor).

    [0093] In the case of short circuits in particular, power is re-established in less than 10 seconds, whilst enabling the total angle to be maintained (which assures stability) with the regulator alone over a wide range of operating points.

    [0094] In the case of voltage dips the voltage at the terminals of the transformer connected to the alternator remains within the limits imposed by EDF in all of the cases examined.

    [0095] In the case of load shedding, the voltage returns to within less than 1% unitary of the final values in less than 10 seconds.

    [0096] In the case of a frequency drop, the return to within 1% unitary of the normal value takes less than 8 seconds, without the voltage error with respect to the setpoint exceeding 4% unitary.

    [0097] The return times, in particular to within 1% of the required value, confirm the fast and effective elimination of oscillations.

    [0098] The method thus enables reduction of the disturbances to the state of the system. By moving the real state towards the ideal state, the method assures system stability that then depends only on the accuracy of the model used and the precision of the measurements.


    Claims

    1. A method for automatic regulation of a system in which a plurality of parameters characteristic of the system are measured and in which at least one control parameter (u) is applied as a function of the measured parameters (y), wherein

    - a nominal operating point of the system is chosen,

    - a nominal model (Mn) describing the system at this nominal operating point is determined, and estimated characteristic output parameters (yr) corresponding to measurable characteristic parameters (y) are determined from said nominal model,

    - the errors between at least one of the measured characteristic output parameters (y) and at least one of the estimated characteristic output parameters (yr) are determined, and

    - at least one control parameter of the system is applied as a function of the errors determined to reduce the error between at least one of the estimated characteristic output parameters (yr) and the corresponding at least one measured value of the measured characteristic output parameters (y), wherein the method is characterized in

    - including a supplementary step of integration of the difference between the command or control parameter (u) and its saturated value (usat), and in that the at least one control parameter of the system applied as a function of the errors determined to reduce the error between the estimated characteristic output parameters (yr) and the measured values of the measurable characteristic output parameters (y) is determined taking into account the integral of the difference between the control parameter (u) and its saturated value (usat).


     
    2. The method claimed in claim 1, characterized in that the nominal model (Mn) is an approximation (Am, Bm, Cm, Dm) of the ideal transfer function of the system.
     
    3. The method claimed in either one of claims 1 or 2, characterized in that the nominal model (Mn) is the linearization (Am, Bm, Cm, Dm) of the ideal transfer function of the system around an operating point.
     
    4. The method claimed in any one of claims 1 to 3, characterized in that said at least one control parameter of the system applied as a function of the errors determined to reduce the error between the estimated characteristic output parameters (yr) and the measured characteristic output parameters (y) is determined by applying optimum feedback to an augmented system derived from the initial system by adding the integral to at least one of the measured predetermined characteristic output parameters.
     
    5. The method claimed in any one of the preceding claims, characterized in that the control parameter of the system that is applied is determined by optimization of an integral criterion.
     
    6. The method claimed in any one of the preceding claims, characterized in that the control parameter of the system that is applied is determined by the Linear Quadratic Gaussian (LQG) optimum control method.
     
    7. The method claimed in any one of the preceding claims, characterized in that it further includes the following steps:

    - a set of representative models ([Mk]) of the possible variations relative to the nominal model (Mn) is determined,

    - the error of the nominal model (Mn) of the system is parameterized by decomposition ([δik]) over all the errors between the models of the set of models ([Mk]) representative of the possible variations and the nominal model (Mn),

    - a given optimization criterion (J) is minimized by varying at least one of the previously obtained parameters ([tiki]) of the error (Δ) relative to the nominal model (M) of the system.


     
    8. The method claimed in claim 7, characterized in that it further includes a supplementary step of optimization of the command (u) with error (Δ) relative to the nominal model (M) fixed by determination of at least one feedback gain (-K(s)).
     
    9. The method claimed in claim 8, characterized in that the steps of minimization of the optimization criterion (J) by varying the parameters of the error (Δ) relative to the nominal model (M) of the system and optimization of the command (u) with error (Δ) relative to the nominal model (M) fixed by determination of at least one feedback gain (-K(s)) are repeated successively in an iterative loop.
     
    10. The method claimed in any one of the preceding claims, characterized in that the system includes an electrical power station alternator (3) connected to an electrical network and its exciter (5).
     
    11. The method claimed in claim 10, characterized in that the state of the system is represented by a state vector (x) that includes the stator voltage (Vs), the rotation speed (ω) of the rotor, the total angle (θ) and an image of the flow in the exciter (efd) (5).
     
    12. The method claimed in claim 10 or 11, characterized in that the set of output magnitudes includes the rotation speed (ω) of the rotor, the active power (Pe) and an approximate value of the mechanical power (Pmec) modeling a main disturbance.
     
    13. The method claimed in at least one of claims 10 to 12, characterized in that said at least one control parameter that is applied includes an approximate value of the mechanical power modeling the main disturbance (Pmec) and the exciter control voltage.
     
    14. A system for automatic regulation of a system in which a plurality of parameters characteristic of the system are measured and in which at least one control parameter is applied as a function of the measured parameters,
    wherein it includes means configured:

    - to choose a nominal operating point of the system,

    - to determine a nominal model (Mn) describing the system at this nominal operating point,

    - to determine estimated characteristic output parameters (yr) corresponding to measurable characteristic output parameters (y) from said nominal model,

    - to determine the errors between at least one of the measured characteristic output parameters (y) and at least one of the estimated characteristic output parameters (yr), and

    - to apply at least one control parameter of the system as a function of the errors determined to reduce the error between at least one of the estimated characteristic output parameters (yr) and the corresponding at least one measured value of the measured characteristic output parameters (u, y),

    wherein the system is characterized in including a supplementary step of integration of the difference between the command or control parameter (u) and its saturated value (usat), and in that the at least one control parameter of the system applied as a function of the errors determined to reduce the error between the estimated characteristic output parameters (yr) and the measured values of the measurable characteristic output parameters (y) is determined taking into account the integral of the difference between the control parameter (u) and its saturated value (usat).
     


    Ansprüche

    1. Verfahren zur automatischen Regulierung eines Systems, bei dem eine Vielzahl von für das System charakteristischen Parametern gemessen wird und bei dem mindestens ein Steuerparameter (u) in Abhängigkeit von den gemessenen Parametern (y) angewendet wird, wobei

    - ein nomineller Betriebspunkt des Systems gewählt wird,

    - ein nominelles Modell (Mn), welches das System an diesem nominellen Betriebspunkt beschreibt, bestimmt wird, und geschätzte charakteristische Ausgangsparameter (yr), die messbaren charakteristischen Parametern (y) entsprechen, aus dem nominellen Modell bestimmt werden,

    - die Fehler zwischen mindestens einem der gemessenen charakteristischen Ausgangsparameter (y) und mindestens einem der geschätzten charakteristischen Ausgangsparameter (yr) bestimmt werden, und

    - mindestens ein Steuerparameter des Systems in Abhängigkeit von den bestimmten Fehlern angewendet wird, um den Fehler zwischen mindestens einem der geschätzten charakteristischen Ausgangsparameter (yr) und dem entsprechenden mindestens einen gemessenen Wert der gemessenen charakteristischen Ausgangsparameter (y) zu reduzieren, wobei das Verfahren dadurch gekennzeichnet ist, dass es

    - einen zusätzlichen Schritt der Integration der Differenz zwischen dem Befehls- oder Steuerparameter (u) und seinem Sättigungswert (usat) einschließt, und dass der mindestens eine Steuerparameter des Systems, der in Abhängigkeit von den bestimmten Fehlern angewandt wird, um den Fehler zwischen den geschätzten charakteristischen Ausgangsparametern (yr) und den gemessenen Werten der messbaren charakteristischen Ausgangsparameter (y) zu reduzieren, unter Berücksichtigung des Integrals der Differenz zwischen dem Steuerparameter (u) und seinem Sättigungswert (usat) bestimmt wird.


     
    2. Verfahren nach Anspruch 1, dadurch gekennzeichnet, dass das nominelle Modell (Mn) eine Approximation (Am, Bm, Cm, Dm) der idealen Transferfunktion des Systems ist.
     
    3. Verfahren nach einem der Ansprüche 1 oder 2, dadurch gekennzeichnet, dass das nominelle Modell (Mn) die Linearisierung (Am, Bm, Cm, Dm) der idealen Transferfunktion des Systems um einen Betriebspunkt ist.
     
    4. Verfahren nach einem der Ansprüche 1 bis 3, dadurch gekennzeichnet, dass der mindestens eine Steuerparameter des Systems, der in Abhängigkeit von den bestimmten Fehlern angewendet wird, um den Fehler zwischen den geschätzten charakteristischen Ausgangsparametern (yr) und den gemessenen charakteristischen Ausgangsparametern (y) zu reduzieren, bestimmt wird, indem eine optimale Rückkopplung auf ein erweitertes System angewendet wird, das von dem ursprünglichen System abgeleitet wird, indem das Integral zu mindestens einem der gemessenen vorbestimmten charakteristischen Ausgangsparameter hinzugefügt wird.
     
    5. Verfahren nach einem der vorstehenden Ansprüche, dadurch gekennzeichnet, dass der Steuerparameter des Systems, der angewendet wird, durch Optimierung eines Integralkriteriums bestimmt wird.
     
    6. Verfahren nach einem der vorstehenden Ansprüche, dadurch gekennzeichnet, dass der Steuerparameter des Systems, der angewendet wird, durch das lineare quadratische Gaußsche (LQG)-Optimierungssteuerverfahren bestimmt wird.
     
    7. Verfahren nach einem der vorstehenden Ansprüche, dadurch gekennzeichnet, dass es ferner die folgenden Schritte einschließt:

    - ein Satz von repräsentativen Modellen ([Mk]) der möglichen Variationen relativ zum nominellen Modell (Mn) wird bestimmt,

    - der Fehler des nominellen Modells (Mn) des Systems wird durch Zerlegung ([δik]) über alle Fehler zwischen den für die möglichen Variationen repräsentativen Modellen des Satzes von Modellen ([Mk]) und dem nominellen Modell (Mn) parametrisiert,

    - ein gegebenes Optimierungskriterium (J) wird minimiert, indem mindestens einer der zuvor erhaltenen Parameter ([tiki]) des Fehlers (Δ) relativ zum nominellen Modell (M) des Systems variiert wird.


     
    8. Verfahren nach Anspruch 7, dadurch gekennzeichnet, dass es ferner einen zusätzlichen Schritt der Optimierung des Befehls (u) mit Fehler (Δ) relativ zum nominellen Modell (M) unter Fixierung durch die Bestimmung mindestens einer Rückkopplungsverstärkung (-K(s)) einschließt.
     
    9. Verfahren nach Anspruch 8, dadurch gekennzeichnet, dass die Schritte der Minimierung des Optimierungskriteriums (J) durch Variieren der Parameter des Fehlers (Δ) relativ zum nominellen Modell (M) des Systems und der Optimierung des Befehls (u) mit Fehler (Δ) relativ zum nominellen Modell (M) unter Fixierung durch Bestimmung mindestens einer Rückkopplungsverstärkung (-K(s)) nacheinander in einer iterativen Schleife wiederholt werden.
     
    10. Verfahren nach einem der vorstehenden Ansprüche, dadurch gekennzeichnet, dass das System einen Wechselstromgenerator (3) eines elektrischen Kraftwerks, der an ein elektrisches Netz angeschlossen ist, und seinen Erreger (5) einschließt.
     
    11. Verfahren nach Anspruch 10, dadurch gekennzeichnet, dass der Zustand des Systems durch einen Zustandsvektor (x) dargestellt wird, der die Statorspannung (Vs), die Drehzahl (ω) des Rotors, den Gesamtwinkel (θ) und ein Bild der Strömung im Erreger (efd) (5) einschließt.
     
    12. Verfahren nach Anspruch 10 oder 11, dadurch gekennzeichnet, dass der Satz von Ausgangsgrößen die Drehzahl (ω) des Rotors, die Wirkleistung (Pe) und einen Näherungswert der mechanischen Leistung (Pmec) einschließt, der eine Hauptstörung modelliert.
     
    13. Verfahren nach mindestens einem der Ansprüche 10 bis 12, dadurch gekennzeichnet, dass der mindestens eine angewandte Steuerparameter einen Näherungswert der mechanischen Leistung, der die Hauptstörung (Pmec) modelliert, und die Erreger-Steuerspannung einschließt.
     
    14. System zur automatischen Regulierung eines Systems, bei dem eine Vielzahl von für das System charakteristischen Parametern gemessen wird und bei dem mindestens ein Steuerparameter in Abhängigkeit von den gemessenen Parametern angewendet wird,
    wobei es Mittel einschließt, die konfiguriert sind zum:

    - Wählen eines nominellen Betriebspunkts des Systems,

    - Bestimmen eines nominellen Modells (Mn), welches das System an diesem nominellen Betriebspunkt beschreibt,

    - Bestimmen geschätzter charakteristischer Ausgangsparameter (yr), die messbaren charakteristischen Ausgangsparametern (y) entsprechen, aus dem nominellen Modell,

    - Bestimmen der Fehler zwischen mindestens einem der gemessenen charakteristischen Ausgangsparameter (y) und mindestens einem der geschätzten charakteristischen Ausgangsparameter (yr), und

    - Anwenden mindestens eines Steuerparameters des Systems in Abhängigkeit von den bestimmten Fehlern, um den Fehler zwischen mindestens einem der geschätzten charakteristischen Ausgangsparameter (yr) und dem entsprechenden mindestens einen gemessenen Wert der gemessenen charakteristischen Ausgangsparameter (y) zu reduzieren,

    wobei das System dadurch gekennzeichnet ist, dass es einen zusätzlichen Schritt der Integration der Differenz zwischen dem Befehls- oder Steuerparameter (u) und seinem Sättigungswert (usat) einschließt, und dass der mindestens eine Steuerparameter des Systems, der in Abhängigkeit von den bestimmten Fehlern angewendet wird, um den Fehler zwischen den geschätzten charakteristischen Ausgangsparametern (yr) und den gemessenen Werten der messbaren charakteristischen Ausgangsparameter (y) zu reduzieren, unter Berücksichtigung des Integrals der Differenz zwischen dem Steuerparameter (u) und seinem Sättigungswert (usat) bestimmt wird.
     


    Revendications

    1. Procédé de régulation automatique d'un système dans lequel une pluralité de paramètres caractéristiques du système sont mesurés et dans lequel au moins un paramètre de contrôle (u) est appliqué en fonction des paramètres mesurés (y), dans lequel

    - un point de fonctionnement nominal du système est choisi,

    - un modèle nominal (Mn) décrivant le système à ce point de fonctionnement nominal est déterminé, et des paramètres de sortie caractéristiques estimés (yr) correspondant aux paramètres caractéristiques mesurables (y) sont déterminés à partir dudit modèle nominal,

    - les erreurs entre au moins l'un des paramètres de sortie caractéristiques mesurés (y) et au moins l'un des paramètres de sortie caractéristiques estimés (yr) sont déterminées, et

    - au moins un paramètre de contrôle du système est appliqué en fonction des erreurs déterminées pour réduire l'erreur entre au moins un des paramètres de sortie caractéristiques estimés (yr) et l'au moins une valeur mesurée correspondante des paramètres de sortie caractéristiques mesurés (y), dans lequel le procédé est caractérisé en ce que

    - il inclut une étape supplémentaire d'intégration de la différence entre le paramètre de commande ou de contrôle (u) et sa valeur saturée (usat), et en ce que l'au moins un paramètre de contrôle du système appliqué en fonction des erreurs déterminées pour réduire l'erreur entre les paramètres de sortie caractéristiques estimés (yr) et les valeurs mesurées des paramètres de sortie caractéristiques mesurables (y) est déterminé en tenant compte de l'intégrale de la différence entre le paramètre de contrôle (u) et sa valeur saturée (usat).


     
    2. Procédé selon la revendication 1, caractérisé en ce que le modèle nominal (Mn) est une approximation (Am, Bm, Cm, Dm) de la fonction de transfert idéale du système.
     
    3. Procédé selon l'une ou l'autre des revendications 1 ou 2, caractérisé en ce que le modèle nominal (Mn) est la linéarisation (Am, Bm, Cm, Dm) de la fonction de transfert idéale du système autour d'un point de fonctionnement.
     
    4. Procédé selon l'une quelconque des revendications 1 à 3, caractérisé en ce que ledit au moins un paramètre de contrôle du système appliqué en fonction des erreurs déterminées pour réduire l'erreur entre les paramètres de sortie caractéristiques estimés (yr) et les paramètres de sortie caractéristiques mesurés (y) est déterminé en appliquant une rétroaction optimale à un système augmenté dérivé du système initial en ajoutant l'intégrale à au moins un des paramètres de sortie caractéristiques prédéterminés mesurés.
     
    5. Procédé selon l'une quelconque des revendications précédentes, caractérisé en ce que le paramètre de contrôle du système, qui est appliqué, est déterminé par optimisation d'un critère intégral.
     
    6. Procédé selon l'une quelconque des revendications précédentes, caractérisé en ce que le paramètre de contrôle du système, qui est appliqué, est déterminé par la méthode de commande quadratique linéaire gaussienne (LQG) optimale.
     
    7. Procédé selon l'une quelconque des revendications précédentes, caractérisé en ce qu'il comprend en outre les étapes suivantes :

    - un ensemble de modèles représentatifs ([Mk]) des variations possibles par rapport au modèle nominal (Mn) est déterminé,

    - l'erreur du modèle nominal (Mn) du système est paramétrée par décomposition ([δik]) sur toutes les erreurs entre les modèles de l'ensemble de modèles ([Mk]) représentatifs des variations possibles et du modèle nominal (Mn),

    - un critère d'optimisation donné (J) est minimisé en faisant varier au moins l'un des paramètres précédemment obtenus ([tiki]) de l'erreur (Δ) par rapport au modèle nominal (M) du système.


     
    8. Procédé selon la revendication 7, caractérisé en ce qu'il inclut en outre une étape supplémentaire d'optimisation de la commande (u) avec erreur (Δ) par rapport au modèle nominal (M) fixée par détermination d'au moins un gain de rétroaction (-K(s)).
     
    9. Procédé selon la revendication 8, caractérisé en ce que les étapes de minimisation du critère d'optimisation (J) en faisant varier les paramètres de l'erreur (Δ) par rapport au modèle nominal (M) du système et d'optimisation de la commande (u) avec l'erreur (Δ) relative au modèle nominal (M) fixée par détermination d'au moins un gain de rétroaction (-K(s)) sont répétées successivement dans une boucle itérative.
     
    10. Procédé selon l'une quelconque des revendications précédentes, caractérisé en ce que le système inclut un alternateur de station d'alimentation électrique (3) connecté à un réseau électrique et à son excitateur (5).
     
    11. Procédé selon la revendication 10, caractérisé en ce que l'état du système est représenté par un vecteur d'état (x) qui comprend la tension de stator (Vs), la vitesse de rotation (ω) du rotor, l'angle total (θ) et une image du flux dans l'excitateur (efd) (5).
     
    12. Procédé selon la revendication 10 ou 11, caractérisé en ce que l'ensemble des grandeurs de sortie inclut la vitesse de rotation (ω) du rotor, la puissance active (Pe) et une valeur approximative de la puissance mécanique (Pmec) modélisant une perturbation principale.
     
    13. Procédé selon au moins l'une des revendications 10 à 12, caractérisé en ce que ledit au moins un paramètre de contrôle, qui est appliqué, inclut une valeur approximative de la puissance mécanique modélisant la perturbation principale (Pmec) et la tension de contrôle de l'excitateur.
     
    14. Système de régulation automatique d'un système dans lequel une pluralité de paramètres caractéristiques du système sont mesurées et dans lequel au moins un paramètre de contrôle est appliqué en fonction des paramètres mesurés,
    dans lequel il inclut un moyen configuré pour :

    - choisir un point de fonctionnement nominal du système,

    - déterminer un modèle nominal (Mn) décrivant le système à ce point de fonctionnement nominal,

    - déterminer des paramètres de sortie caractéristiques estimés (yr) correspondant aux paramètres de sortie caractéristiques mesurables (y) à partir dudit modèle nominal,

    - déterminer les erreurs entre au moins l'un des paramètres de sortie caractéristiques mesurés (y) et au moins l'un des paramètres de sortie caractéristiques estimés (yr), et

    - appliquer au moins un paramètre de contrôle du système en fonction des erreurs déterminées pour réduire l'erreur entre au moins l'un des paramètres de sortie caractéristiques estimés (yr) et l'au moins une valeur mesurée correspondante des paramètres de sortie caractéristiques mesurés (u, y),

    dans lequel le système est caractérisé en ce qu'il inclut une étape supplémentaire d'intégration de la différence entre le paramètre de commande ou de contrôle (u) et sa valeur saturée (usat), et en ce que l'au moins un paramètre de contrôle du système appliqué en fonction des erreurs déterminées pour réduire l'erreur entre les paramètres de sortie caractéristiques estimés (yr) et les valeurs mesurées des paramètres de sortie caractéristiques mesurables (y) est déterminé en tenant compte de l'intégrale de la différence entre le paramètre de contrôle (u) et sa valeur saturée (usat).
     




    Drawing

















    Cited references

    REFERENCES CITED IN THE DESCRIPTION



    This list of references cited by the applicant is for the reader's convenience only. It does not form part of the European patent document. Even though great care has been taken in compiling the references, errors or omissions cannot be excluded and the EPO disclaims all liability in this regard.

    Patent documents cited in the description