This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication.

Multiple Model Predictive Control for Wind Turbines With Doubly Fed Induction Generators
M. Soliman,

O.P. Malik, LFIEEE D.T. Westwick , MIEEE

Abstract A multivariable control strategy based on model predictive control techniques for the control of variable speed variable pitch wind turbines is proposed. The proposed control strategy is described for the whole operating region of the wind turbine, i.e. both partial and full load regimes. Pitch angle and generator torque are controlled simultaneously to maximize energy capture, mitigate drive train transient loads and smooth the power generated while reducing the pitch actuator activity. This has the effect of improving the efficiency and the power quality of the electrical power generated, and increasing the life expectancy of the installation. Furthermore, safe and acceptable operation of the system is guaranteed by incorporating most of the constraints on the physical variables of the WECS in the controller design. In order to cope with nonlinearities in the WECS and continuous variations in the operating point, a multiple model predictive controller is suggested which provides acceptable performance throughout the whole operating region. Index Terms Wind power generation, Predictive control, Doubly fed induction generators.

above and below the cut out wind speed, , the turbine is operating in the full load regime. In this region the control system should regulate the output power and the generator speed at their rated values. The design of WECS control systems is not a straight forward task. One of the main reasons for this is that the system is a multiple-input multiple-output (MIMO) system with strongly coupled variables. Furthermore, the system non-linearity, the stochastic variations of the input power and the presence of physical constraints on the systems variables render the control design task more difficult.
1.5 Partial Load Full Load

P (MW)

1 0.5 0

v ci
0 5 10

v
15 20 25

co

30

v (m/s)
40 Partial Load 30 Full Load

I. INTRODUCTION

20 10

ONTROL systems play a very important role in modern Wind Energy Conversion Systems (WECSs). A well designed WECS control system enables efficient energy generation, good power quality and the alleviation of aerodynamic and mechanical loads resulting in increased life of the installation. Consequently, such a control system will have a direct impact on the cost of energy produced [1], [2]. In general, a variable speed variable pitch WECS has two operating regions, as shown in Fig. 1 [1]. The partial load regime includes all wind speeds between the cut in wind speed, , and the rated wind speed, (wind speed at which the system rated power is achieved). In this region, the control system should adjust the turbine rotor speed such that maximum energy is extracted. When the wind speed is
This work was supported by the Alberta Ingenuity Fund (File # 200800408). M. Soliman is is currently working toward the Ph.D. degree in Electrical Engineering Department, University of Calgary, Calgary, AB, Canada T2N 1N4 (phone: 403-210-9515; fax: 403-282-6855; e-mail: msoliman@ucalgary.ca). O.P. Malik is with Electrical Engineering Department, University of Calgary, Calgary, AB, Canada T2N 1N4 (e-mail: maliko@ucalgary.ca). D.T. Westwick is with Electrical Engineering Department, University of Calgary, Calgary, AB, Canada T2N 1N4 (e-mail: dwestwic@ucalgary.ca).

v
0 0

ci

v
5

(o) t (rpm)
15 20 25

co

10

30

v (m/s)

Fig. 1. Ideal power curve for a WECS.

Many control techniques have been proposed to control WECSs in the partial load regime [1], [4]-[8]. The design of classical Proportional Integral (PI) controllers is described in [4]-[5] and [8]. To cope with the system non-linearity, the use of a gain-scheduling Linear Quadratic Gaussian (LQG) controller is proposed in [6]. A gain scheduled controller is suggested in [1] and [7]. Many papers focusing on the control of variable speed variable pitch WECSs operating in the full load regime have appeared recently [8]-[10]. Most of the work reported ignores the multivariable nature of the problem [8]-[9]. Classical PI controllers are used in [8]. A PI controller in the power loop and an adaptive self-tuning regulator in the speed loop are proposed in [9]. A multivariable gain scheduled controller is proposed in [10]. The use of Model Predictive Control (MPC) techniques to control the WECS in the full load regime is proposed in [11]. The presence of two control regions with different control structures requires the ability to switch between these

Copyright (c) 2011 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

2 controllers. Several recent studies indicate that when wind speed fluctuates around its rated value, undesirable drive train transient loads and power overshoots can occur [1]. This paper is motivated by the desire to develop an overall control strategy that can work in both partial and full load regimes. Furthermore, as discussed in [1] and [10], recognizing the multivariable nature of the problem and designing a MIMO controller will lead to much superior performance as compared to the decentralized approach commonly used in the literature. To the knowledge of the authors, the only work that describes the design of a multivariable controller that can work in both partial and full load regimes is found in [1],[12], where a multivariable gain scheduled controller is proposed. A new control strategy based on MPC techniques is proposed to control variable speed variable pitch WECSs in both partial and full load regimes. The paper focuses on the Doubly Fed Induction Generator (DFIG) configuration since it is the most popular type used today. The proposed strategy is a multivariable method that uses the full capability of the system to obtain the desired performance in the whole operating region of the WECS, while keeping the system variables within safe limits. Some preliminary work related to the proposed MPC control strategy was presented in [11] and [13]. This paper extends [11] and [13] by designing a controller for the whole WECS operating region (partial load and full load regimes). Furthermore, a systematic way to damp drive torsional torque oscillations is proposed and more detailed simulation studies, including the DFIG dynamics when interconnected to a small distribution power network, are carried out. II. OVERVIEW OF DFIG BASED WIND TURBINES
WECS Optimization * g* PDFIG Turbine * * * Qs QGC Vdc Controller * Tg*
RSC Controller GSC Controller

is connected to a back-to-back voltage source converter, as shown in Fig. 2. The back-to-back converter consists of the rotor side converter (RSC) and the grid side converter (GSC), connected with a direct current (DC) link. B. Control system hierarchy Due to large differences in the time scales of the electrical and mechanical dynamics, the DFIG based WECS has a multilayer control structure as shown in Fig. 2. The highest control level is the WECS optimization level. At this control level, a Maximum Power Point Tracking (MPPT) algorithm [2] is used to calculate the generator speed set point, , so that the energy conversion efficiency is maximized in the partial load regime. In the full load regime, this control level generates , and the DFIG output power set point, , so that the generator speed and power are regulated to their rated values. The second control level contains the turbine controller. This control level supervises the pitch control system and the generator controller. The task of the turbine controller is to control the generator speed and output power to follow their desired values. This is achieved by manipulating the pitch angle set point, , and the generator torque set point, . The third control level contains the generator controller. This controller regulates the generator torque, the stator reactive power, the GSC reactive power and the DC link voltage to follow their set points , , and , respectively, by controlling the RSC and the GSC. This paper will focus on the design of the turbine controller using model predictive control techniques. III. MODELING OF DFIG BASED WECSS A model of the entire WECS can be structured as several interconnected subsystems as shown in Fig. 3 [1]. Details of individual blocks are given in Appendix A [1]-[5] and [14].
Wind speed simulator v Pitch actuator system

Generator Controller
Smoothing inductor

Aerodynamics Tt Drive Train model Tg t g

vr* Gear box

RSC DC link

vC*
GSC
T g*

Electrical Subsystem
feedback

Generator Controller

DFIG System

PDFIG

Wind turbine

PDFIG QDFIG WRIG

Grid
iDFIG vs Power system Grid

Fig. 3. WECS model.

Fig. 2. DFIG based WECS and its control system.

A. Effective wind speed model A realistic wind speed model assessment of the proposed control the effective wind speed, , composed of two components, is necessary for the strategy. In this paper, described by (1), is the slowly varying

A. The DFIG system A DFIG based WECS consists of a three blade wind turbine rotor coupled to a Wound Rotor Induction Generator (WRIG) through a gear box. The rotor winding of the WRIG

Copyright (c) 2011 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

3 component, , and the turbulent component . The turbulent component, , is obtained by filtering a white noise signal by three cascaded filters , and . The filter is used to approximate the von Karman spectrum [1]. To include the effect of the blades rotation in a non-uniform wind speed field, a low pass spatial filter, , is used to attenuate the high-frequency components of the turbulence, and the rotational sampling filter, , is used to amplify those components with frequencies close to triple the turbine speed (3P frequency). Details of these filters are given in [1], [15]-[16]. (1) B. Simplified WECS model For turbine controller design, it is important to use simple models that capture the relevant dynamics of the system. Fortunately, the dynamics of the electrical subsystem are much faster than the turbine dynamics and simple models can be used to represent the electrical dynamics. In this paper, a first order model, given in (2)-(3), is used. (2) (3) Here, , and are the generator torque, time constant and efficiency, respectively. Despite this simplification, the resulting simplified WECS model described in (A.1)-(A.8) and (2)-(3) is non-linear. Linearizing the turbine torque equation in (A.3) yields: (4) (5) The symbol is used to represent the deviation of a variable from its operating point value while the over bar in denotes the value of a variable at the operating point. Since the steady state values and are dependent on , as shown in Fig. 1, the WECS operating point is completely defined by [1].The linearized state space model of the WECS can be written as (6)-(8). (6) Here, state vector, is the is the control input and is the measured output. It can be seen that the system is MIMO and that the system dynamics vary when the average wind speed varies. IV. CONTROL PROBLEM DESCRIPTION There are three main control objectives of WECS control systems. First, the control system should maximize the energy capture. Second, transient loads should be minimized, consequently maximizing the life expectation of the installation. Third, voltage flicker emissions should be reduced thus enhancing the power quality [17]. In the partial load regime, the primary objective is to control the turbine rotor speed to maximize the turbine's aerodynamic efficiency. This is achieved by manipulating and fixing at its optimal value (usually very close to zero). The main control challenge is to design a controller that can maximize the conversion efficiency while minimizing transient loads. It was found that there is a conflict between these two objectives and a trade-off must be decided [6]. In the full load regime, the main control objective is to regulate both the DFIG output power, , and the generator speed, , at their rated values and , respectively. One of the main challenges in this regime is the presence of severe fluctuations in the turbine power, , caused by erratic variations in the wind speed. Fluctuations in can lead to large variations in the drive train torsional torque, , and in the electric power supplied to the grid. These, in turn, can cause a reduction in the life time of the WECS components and flicker problems [8]-[9]. A common challenge in both partial and full load regimes is the presence of non-linearity in the system dynamics and the continuous variation of the operating point. Another challenge is the presence of cyclic aerodynamic torque variations at triple the speed of rotation of the wind turbine blades. This frequency is normally referred to as the 3P frequency [17]. This phenomenon, known as the rotational sampling effect [1], is caused by the blades crossing a nonuniform wind field. These cyclic torque variations can increase dynamic loads and increase the voltage flicker. Finally, the control system should realize all the control objectives while keeping system variables within acceptable limits. For example, there are maximum limits on the amplitude and speed of the pitch servos, the generated power and the turbine speed. Ignoring such constraints during the controller design can lead to performance degradation [18] and, in the worst case, system shut down. V. CLASSICAL WECS CONTROL STRATEGY The classical wind turbine control strategy [12] relies on using a set of two PI controllers to regulate the generator speed in the partial and full load regimes, respectively. In partial load operation, is fixed at zero and is manipulated by the first PI controller so that the generator

(7)

(8)

Copyright (c) 2011 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

4 speed tracks . In the full load regime, is fixed at its rated value, , while is manipulated by the second PI controller to regulate the generator speed at its rated value. In order to take into account variations in the aerodynamics, these PI controllers are generally gain scheduled. The main shortcomings of this classical strategy are explained as follows. In the partial load regime, the PI controller design does not allow fine tuning of the trade-off between energy maximization and transient load minimization [2]. Furthermore, when the system is operating near the rated wind speed, significant power and drive train torsional torque overshoots occur due to switching between the partial load and full load control configurations [1]. Finally, in the full load regime, using the pitch actuator alone to regulate the generator speed can cause large pitch activity and severe power fluctuations. VI. PROPOSED CONTROL STRATEGY A. Review of Model Predictive control Model-based Predictive Control (MPC) has been successfully used in many industrial applications due to its ability to handle MIMO control problems with constraints on the system variables [18]-[20]. The basic idea behind MPC can be explained as follows. First, a dynamic model of the system to be controlled and the physical constraint specifications on system variables must be known. At each sampling time, the future outputs of the system are predicted within a predefined prediction horizon. The set of future control signals is calculated by solving a constrained optimization problem involving the system constraints and a user defined performance index that reflects the system performance. The first input in the optimal sequence is then sent to the system, and the entire calculation is repeated at subsequent control intervals with shifted prediction horizons. This idea is illustrated in Fig. 4.
Output
Prediction Horizon

B. Multiple Model Predictive Control (MMPC) for variable speed variable pitch WECSs The proposed MMPC based WECS control strategy is shown in Fig. 5. The main idea is to use a multivariable MPC controller at the turbine control level to control the WECS behavior by manipulating both and . WECS constraints, such as limits on the pitch angle magnitude, pitch angle rate, the generated power and the turbine speed, are explicitly incorporated in the MPC controller. To cope with WECS non-linearities and the continuous variation in the operating point, the whole operating region of the WECS is divided into operating sub-regions with linearized models that adequately represent the local system dynamics within each sub-region. A linear MPC controller based on each model is designed. Finally, a criterion by which the control system switches from one controller to another as operating conditions change is defined. This approach is known in the literature as MMPC [21]- [22]. The main components of the proposed MMPC strategy are the prediction model bank, the optimization solver, the state estimator and a model switching criterion [18], [20].
MMPC
Scheduling Signal Model Bank

* PDFIG

Ai, Bui, Ci
MPC

*
WECS

PDFIG

WECS Optimization

g g*

i x i (k | k ) xd (k | k )
Estimator Bank

Tg*

Fig. 5. Proposed control strategy using MMPC.

Reference Constraints Time Input Constraints Time

k k+1 k+Np

1) Prediction model bank A model bank (9), consisting of linearized models that represent the WECS dynamics in the whole operating region shown in Fig. 1 must be available.

(9) For the case of the WECS described in Section III, the control input vector , the state vector and the controlled measurement vector of model in (9) at the sampling instant are defined in (10).

Fig. 4. MPC concept.

The main drawback of using MPC controllers is the requirement to solve a quadratic programming (QP) problem on line. This has restricted the use of MPC to applications with slow dynamics. However, due to the advances in computational power and in optimization algorithms, MPC can be used now for systems with fast dynamics [19].

(10) In the case of the WECS, one of the control objectives besides the tracking/regulation of the measurements is to damp high frequency oscillations in the drive train torsional

Copyright (c) 2011 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

5 torque, . In the optimal control framework, this is achieved by defining an unmeasured controlled signal, , as in (10), and by including this signal in the objective function of the optimization problem [23]. The fictitious unmeasured disturbance is used to represent the effect of actual unmeasured disturbances and it is modeled as the output of the system (11), where is assumed to be Gaussian white noise. (11) The matrices , , and in (9) with sampling period, , are computed from the continuous plant model in (6) at different wind speeds, , using (12)-(13), where and denote the third row in and , respectively. (12) (13) Combining (9) and (11), the augmented prediction model bank used in the MPC formulation is given by: , (20) Using the MPC formulation in (15)-(18), there are six weights, , , , , and , that can be tuned by the designer to reach the desired compromise between different control objectives. Details of weight selection for partial and full load regimes are provided in Section VI.C.1. 3) State estimation In this paper, the state estimates are computed using an observer bank consisting of state observers in (20)-(21)

(14)

2) Optimization problem Assuming knowledge of the estimates of the plant states and disturbance states given the data up to time , the MMPC controller solves the quadratic optimization problem given by (15)-(18).

(21) The gain, , is designed using Kalman filtering techniques [20] based on model in (14). 4) Bumpless switching between different MPC controllers The switching between different MPCs is based on the value of a scheduling signal as shown in Fig. 5. In the case of a WECS, the scheduling signal can be the generator speed and pitch angle or an average wind speed estimate which can be determined online by filtering the wind speed measured by an anemometer located at the wind turbine [10], [12]. Due to wind speed fluctuations, it is important to ensure bumpless switching between different MPC controllers. This is ensured by (i) using an MMPC algorithm that calculates only the control increments, and , and (ii) by continuously updating the internal state of all estimators in the estimator bank. This reduces the transients in the state estimates when switching between MCP controllers. C. MMPC tuning for variable speed variable pitch WECSs 1) MMPC Weight selection In the proposed MMPC strategy, the whole operating region is partitioned into sub-regions. For sub-regions corresponding to partial load operation, the weights and in (15) should be set to large values to force the pitch angle set point to be fixed at zero. Since the objective is to force the generator speed to track its set point, the weight should be set to zero while the weights and in (15) should be selected to achieve the desired trade-off between energy maximization and drive train transient load minimization. The desired damping of oscillations in the drive train torsional torque can be achieved by tuning . For sub-regions corresponding to partial load operation near the rated wind speed, the weights , , and should be selected similar to those used in sub-regions corresponding to partial load operation. However, the weights and should be reduced in comparison with subregions corresponding to partial load operation and the constraint (18-e) should be replaced with (22). This selection allows the pitch system to be activated when required to prevent the power from exceeding its rated value when the wind speed fluctuates near the rated wind speed.

(15) Subject to: Prediction model equations in (14) , (16) (17) (18-a) (18-b) (18-c) (18-d) (18-e) is the control

Here,

is the prediction horizon,

horizon, is the tracking error, and , are defined in (10). The control move, , is defined as . The symbol denotes a weighted norm and denotes the maximum (minimum) dynamical limit of . Knowing and the actual physical limits, and , the limits and can be calculated. The weights , and are defined as: (19) where .

Copyright (c) 2011 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

6 Consequently, power and drive train twist torque overshoots above their rated values are eliminated. (22) For sub-regions corresponding to the full load regime, the weight should be set to zero to allow the pitch angle to take any required value. The weights , , and are tuned to achieve the desired trade-off between generator speed regulation, power smoothing, drive train transient loads reduction and pitch angle activity, respectively. 2) Disturbance model selection The disturbance model in (11) is chosen to guarantee offset free tracking for the WECS [24]. This property ensures that when the system reaches steady state, the controlled variables are equal to their desired set points. Following the guidelines in [24], two integrated white noise unmeasured disturbances, , entering at the inputs are assumed. This can be achieved by using (23). (23) This selection ensures offset free tracking for the system described in [24], where the controlled output used in the MPC objective function is a linear combination of the measurements. However, this is not the case in this paper due to the presence of a term containing the unmeasured controlled output, , in (15). Appendix C provides a proof that at steady state, the MPC controller used in this paper is equivalent to the one in [24] and therefore, offset free tracking is guaranteed for the proposed MPC strategy. 3) Partitioning the whole operating region into operating sub-regions The proposed MMPC control strategy is based on partitioning the whole operating region into operating sub-regions as shown in (24). The main shortcoming of this approach is that it requires the knowledge of the effective wind speed experienced by the rotor blades. This fictitious speed cannot be measured and it is very difficult to estimate. The approach that is used in this paper, shown in Fig. 6, is similar to the one described in [4], [5], [8] and [25]. An experimental validation of this approach on a DFIG test bench is provided in [4]. This approach does not require knowledge of the effective wind speed; and is calculated using (25), where is an estimate of the turbine torque. The turbine torque estimator described in [4] and [25] is used in this paper. Using (A.3), it easy to show that when the turbine is operating at the optimal tip speed ratio, the relation in (25) is satisfied. ,
Tt kopt

(25)
* g
NM

Tt
Fig. 6. MPPT algorithm.

Estimator

g
PDFIG

Remark 1: The generator torque scaled by the gear box ratio can be used in (25) as a rough estimate for the turbine torque. However, this approach ensures optimal operation of the wind turbine in steady state only. That is why a turbine torque estimator is commonly used [4], [8] and [25]. VIII. SIMULATION RESULTS A. Simulation set-up In this section, the performance of the proposed control strategy is compared with the classical gain scheduled PI control strategy described in Section V. For a fair comparison, the same MPPT algorithm, shown in Fig. 6, is used with both controllers in all simulations. Using the algorithm proposed in [11], the whole operating region, with and given in Appendix B, is partitioned into six operating sub-regions defined in Table I. For each of these sub-regions, a linearized model (9) at is assumed to be available. Using these models, six MPC controllers are designed following the guidelines in Section VI.C.1, resulting in the settings listed in Table I. The other MMPC parameters are chosen as: , , . (26) The system under study, shown in Fig. 7, assumes that a 1.5 MW DFIG wind turbine is connected to a radial system [26]. The Short Circuit Ratio (SCR) at the Point of Common Coupling (PCC) is 13.3 MVA and the short circuit impedance angle is 61 . In all simulations presented in this section, the designed controllers are tested on the non-linear model described in Appendix A, with data given in Appendix B. The speed, torque and power signals have been normalized based on the per unit system described in [3].

(24) Special care must be taken when partitioning the whole operating region. In general, increasing the number of partitions and, consequently, reducing the range of each subregion will enhance the linear approximation and the prediction accuracy. This comes at the cost of increasing the controller complexity and the computational burden. In this paper, the algorithm proposed in [11] will be used to generate the partitions in (25). The algorithm is based on partitioning the whole operating region of the WECS such that the prediction mismatch between different linear models within each partition is bounded by certain value [11]. VII. MPPT ALGORITHM Many MPPT algorithms can be used to calculate that is sent to the turbine controller during partial load operation [2]. One alternative is to adjust to change linearly with the effective wind speed such that the tip speed ratio is always kept at its optimal value, [1], [6], [12] and [13].

Copyright (c) 2011 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

7
TABLE I
MMPC SUB-REGIONS AND CONTROLLER DATA

Partial Load Full Load 1 2 3 4 5 6 [4, [8.7, [13, [18.4, [21.8, , [11,13) 8.7) 11) 18.4) 21.8) 26] m/s 6.4 10 12 15.7 20 24 1 2 2 Same as sub-region 3 0 0 5 Same as sub-region 3 0.05 0.05 0.2 Same as sub-region 3 1 1 10 Same as sub-region 3 1000 0.15 0.2 Same as sub-region 3 1000 0.01 0 Same as sub-region 3 Remark 2: It was found that the MMPC calculations required less than 25 ms (about 50% of the sampling time) in simulations carried on a 1.66 GHZ dual core. Remark 3: The PI controllers of the classical strategy were tuned such that the settling time of the closed loop system is similar to that obtained using the MMPC controller. B. Performance measures In order to compare different control strategies, there must be some performance indicators that reflect the performance of the closed loop system. In the partial load regime, controller tracking performance is measured using the average power produced during the entire simulation. To measure the flicker emission, a digital flickermeter is implemented in Matlab [25], based on the IEC 61000-4-15 standard [26], to calculate the short-term flicker severity at the PCC bus shown in Fig. 7. Finally to compare the pitch activity, the standard deviation of the pitch rate will be used.
1 T12 T36 6 L6 T47 7 L7 DFIG T48 9 8 L9 WECS T59 2 3 4 PCC 5

the MMPC control strategy is used. Table II indicates that the MMPC strategy provides around 45% reduction in the flicker emission in comparison to the classical strategy.
7 6.5 6 5.5

v, m/s

10

20 30 time, s

40

50

0.8 0.75 0.7 0.65 0 10

MPC PI g
*

g, p.u.

20 30 time, s

40

50

0.4

MPC PI

Tg, p.u.

0.3 0.2 0.1

10

20 30 time, s

40

50

0.3

MPC PI

PDFIG, p.u.

0.25 0.2 0.15

Grid

Fig. 7. Power system studied.

C. Simulation results 1) Partial load with variable-speed operation (low wind speed) A simulation of ten minutes of partial load WECS operation was performed. The first fifty seconds of this simulation are shown in Fig. 8. It can be seen that both the MMPC and the classical controller provide similar tracking performance of the generator speed. Calculating the average power in both simulations, as shown in Table II, reveals that both controllers have similar power production. Furthermore, it can be observed in Fig. 8 that and fluctuate regardless of the type of control strategy used. This is due to the continuous variation in to track the MPP. However, the amplitudes of these oscillations decrease when

20 30 40 time, s Fig. 8. Simulation results for low wind speeds. TABLE II LOW WIND SPEEDS STATISTICS

0.1

10

50

MPC PI MPC/PI 1 0.2116 0.2116 0.0884 0.1596 0.55 2) Partial load operation at near rated wind speed (medium wind speed) A simulation result for partial load operation near the rated wind speed (11m/s) is shown in Fig. 9. It can be seen that power and drive train torque overshoots occur when using the classical control strategy. The MMPC eliminated these overshoots and the generator power and speed stay within the rated values given in Appendix B. However, this is achieved by increasing the pitch activity. Statistics given in Table III confirm these observations. It can be seen in the

Quantity AVG( )

Copyright (c) 2011 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

8 third column in Table III that the MMPC controller reduces the maximum power, the maximum drive train torque and the flicker emission by about 12%, 8% and 21%, respectively, when compared to the classical PI strategy.
13 12
v, m/s

11 10 9 8 0 100 200

3) Full load operation (High wind speed) A simulation result for full load operation of the WECS is shown in Fig. 10. It can be observed that when using the MMPC controller, the power fluctuations are much reduced. The price of this reduction is an increase in the generator speed fluctuations in comparison with the classical control strategy. Furthermore, Table IV shows about 32% reduction in the pitch activity and 40% reduction in the flicker emission are obtained when using the MMPC controller.
26 24
v, m/s

300 time, s

400

500

600

22 20 18 16

1.2

g, p.u.

1.1 1 0.9
MPC PI g
*

100

200

300 time, s

400

500

600

100

200

g, p.u.

300 time, s

400

500

600
1.2

0.8
Ttw , p.u.

0.6 0.4 0.2

MPC PI

1.15

MPC PI 0 100 200 300 time, s 400 500 600

0.91

100

200

PDFIG, p.u.

300 time, s

400

500

600

0.905 0.9 0.895 0.89

MPC PI

1
PDFIG, p.u.

0.8 0.6 0.4 0.2 0 100 200

MPC PI

100

200

300 time, s

400

500

600

35

300 time, s

400

500

600

30
o

25 20 15 10 0 100 200 MPC PI 300 time, s 400 500 600

MPC

2
o

PI

1 0 -1

,
0 100 200 300 time, s 400 500 600

Fig. 10. Simulation results for high wind speeds. TABLE IV HIGH WIND SPEEDS STATISTICS

Quantity STD( ) STD( )

Fig. 9. Simulation results for medium wind speeds. TABLE III MEDIUM WIND SPEEDS STATISTICS

Quantity STD( ) Max( ) Max( )

MPC 0.1102 0.2412 0.7812 0.9005

PI 0.1388 0.0959 0.8460 1.0191

MPC/PI 0.79 2.51 0.92 0.88

MPC 0.0068 1.1081 0.0109

PI 0.0114 1.6198 0.0028

MPC/PI 0.596 0.684 3.89

D. Robustness of the proposed strategy To investigate the robustness of the proposed MMPC to parametric uncertainties, changes of the nominal

Copyright (c) 2011 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

9 values of the parameters , , and are assumed. The response of the uncertain systems is compared to the nominal response, after a step change in the generator speed command , in Fig. 11. It can be seen that the MMPC controller is robust against parametric uncertainties. Similar results are obtained when testing the MMPC in the full load regime. Results are omitted due to space limitation. Remark 5: If the uncertainty ranges of the model parameters are known, a robust MPC controller can be designed systematically using the approach proposed in [27].
0.88 0.86 0.84

2) Aerodynamic system (A.3) (A.4) (A.5) Here, is the turbine (aerodynamic) torque; blade length; is the power coefficient; the tip speed ratio and 3) Drive train model is the is

is the speed of the low speed shaft. (A.6) (A.7)

wg, p.u.

0.82 0.8 0.78 0.76 0.74 0 2 4 time, s 6

* g
Nominal Uncertain 8 10

(A.8) Here, and are the inertia of the turbine and the generator, respectively; is the drive train torsional torque; is the gear ratio; , are the shaft stiffness and damping coefficients, respectively. 4) DFIG model
a) Induction generator model

Fig. 11. Closed loop step response comparison of the uncertain systems and the nominal system controlled by the MMPC controller.

IX. CONCLUSIONS A multivariable control strategy based on MPC techniques is proposed to control variable speed variable pitch WECSs over their full operating ranges. In the partial load regime, the MMPC controller can be designed to provide the desired tradeoff between energy maximization and reduction of the drive train torsional torque. Near the rated wind speed, power and drive train torsional torque overshoots are eliminated and flicker emissions can be significantly reduced. In the full load regime, the MMPC controller uses both the pitch angle and the generator torque to regulate the generator power and speed. This reduces pitch activity, smoothes the generated power and reduces flicker emissions. Furthermore, the MMPC controller provides the desired WECS performance while keeping the system variables within safe operating limits. Performance of the MMPC controller is compared with the classical gain scheduled PI control strategy. Simulation results show superiority of the proposed strategy over the whole operating region of the WECS. Experimental validation of the proposed strategy is currently under investigation. APPENDICES A. Appendix A: DFIG based WECS model [1]-[5] 1) Pitch actuator system (A.1) Here, , (A.2) is the time constant of the pitch system and is the maximum (minimum) limit of . (A.9)

(A.10) (A.11) In (A.9)-(A.11), , , and denote voltages, currents, flux linkages, resistances and inductances, respectively. The subscripts and denote the direct and quadrature axis components, respectively. The subscripts and denote generator stator and rotor quantities, respectively. and are the generator synchronous speed and the number of pole pairs of the generator, respectively.
b) Model of the GSC connection to the grid

(A.12) In (A.12), , , and are the grid side filter resistance, inductance, the direct and quadrature axis voltage at the GSC terminals respectively.
c) DC link model

(A.13) (A.14) Here, , and denote the dc link capacitance, the active power of the GSC and of the rotor, respectively. 5) Generator controller Detailed description of the generator controller based on vector control techniques is given in [14].

Copyright (c) 2011 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

10 B. Appendix B: DFIG based WECS data 1) Wind turbine/drive train/pitch system System rated power = 1.5 MW(0.9 p.u.); Rated turbine speed = 23 rpm (1.2 p.u.); Min/Max turbine speed = 9.5/25 rpm; = 35 m; / / = 4/11/26 m/s; = 62.6; Turbine inertia constant = 3 s; Generator inertia constant = 0.5 s; Shaft stiffness = 0.5 pu; Shaft damping = 0.01 pu; / = 0/45 o; / = -10/10 o/s; = 0.1 s. 2) DFIG system Rated generator apparent power = 1.5/0.9 MVA; = 60 Hz; = 3; = 0.00706 p.u.; = 0.171 p.u.; = 0.005 p.u.; = 0.156 p.u.; = 2.9 p.u.; = 0.15 / 100 p.u.; = 0.15 p.u.; = ; Rated DC Link Voltage = 1200 V C. Appendix C Theorem: consider a WECS controlled by an MMPC described in Section VI. If the closed loop system is stable and constraints are not active at steady state, then and asymptotically. Proof: The only difference between the MPC controller used in this paper and the one described in [24] is the presence of the unmeasured controller output, , defined (10), in the objective function (15). To simplify notation, the superscript in (9)-(21) is dropped. First, it should be noticed that the system (6) is a nonintegrating process and both and are invertible. By assumption, when the closed loop system reaches steady state, it follows from the observer equations in (20)(21) and the disturbance model matrices (23) that [24]: , (C.1) where , , and are the steady state values of the measurement vector, control input, plant state estimate and disturbance state estimate. Substituting (C.1) and (23) in the prediction model equations (14), (C.2) is obtained. , Using (C.1) and (13) in (C.2) gives: . (C.3) Since is invertible and by using (12), the following results are obtained. , (C.4) From (C.3), (C.4) and since from (8), it can be concluded that at steady state and, therefore, the objective function (15) is equivalent to the one used in [24] and the rest of the proof will be similar to the proof of Theorem 1 in [24]. REFERENCES
[1] F.D. Bianchi, H. Battista and R.J. Mantz, Wind Turbine Control Systems: Principles,Modelling and Gain Scheduling Design. London: Springer-Verlag, 2007. [2] I. Munteanu, A.I. Bratcu and N.A. Cutululis, Optimal Control of Wind Energy Systems. Springer-Verlag, 2008. [3] V. Akhmatov, Induction Generators for Wind Power. Multi-Science Publishing Co. Ltd., 2007 [4] H. Camblong, I. M. de Alegria, M. Rodriguez, and G. Abad, "Experimental evaluation of wind turbines maximum power point tracking controllers," Energy Convers. Manage., vol. 47, pp. 2846-58, 2006. [5] A. D. Hansen, P. Sorensen, F. lov, and F. Blaabjerg, "Control of variable speed wind turbines with doubly-fed induction generators," Wind Eng., vol. 28, pp. 411-43, 2004. [6] I. Munteanu, N. A. Cutululis, A. I. Bratcu and E. Ceanga, "Optimization of variable speed wind power systems based on a LQG approach," Control Eng. Pract., vol. 13, pp. 903-12, 07. 2005. [7] F. D. Bianchi, R. J. Mantz and C. F. Christiansen, "Control of variablespeed wind turbines by LPV gain scheduling," Wind Energy, vol. 7, pp. 1-8, 2004. [8] E.A. Bossanyi, "The design of closed loop controllers for wind turbines," Wind Energy, vol. 3, pp. 149-163, 2000. [9] E.B. Muhando, T. Senjyu, A. Yona, H. Kinjo and T. Funabashi, "Disturbance rejection by dual pitch control and self-tuning regulator for wind turbine generator parametric uncertainty compensation," IET Control Theory and Applications, vol. 1, pp. 1431-1440, 2007. [10] F.D. Bianchi, R.J. Mantz and C.F. Christiansen, "Power regulation in pitch-controlled variable- speed WECS above rated wind speed," Renewable Energy, vol. 29, pp. 1911-22, 2004. [11] M. Soliman, O. P. Malik, and D. Westwick, "Multiple Model MIMO Predictive Control for Variable Speed Variable Pitch Wind Energy Conversion Systems," IET Renewable Power Generation, 2010. (accepted). [12] K. Z. Ostergaard, J. Stoustrup and P. Brath, "Linear parameter varying control of wind turbines covering both partial load and full load conditions," Int. J. Robust Nonlinear Control, vol. 19, pp. 92-116, 2009. [13] M. Soliman, O. P. Malik, and D. Westwick, "Multiple Model MIMO Predictive Control for Variable Speed Variable Pitch Wind Turbines," in 2010 American Control Conference, Baltimore, MD, USA, 2010. [14] R. Pena, J. C. Clare and G. M. Asher, "Doubly fed induction generator using back-to-back PWM converters and its application to variable-speed wind-energy generation," IEE Proc. Electric Power Applications, vol. 143, pp. 231-41, 05. 1996. [15] T. Petru and T. Thiringer, "Modeling of wind turbines for power system studies," IEEE Transactions on Power Systems, vol. 17, pp. 1132-9, 2002. [16] H. Camblong, M. R. Vidal, and J. R. Puiggali, "Principles of a simulation model for a variable-speed pitch-regulated wind turbine," Wind Engineering, vol. 28, pp. 157-75, 2004. [17] T. Sun, Z. Chen and F. Blaabjerg, "Flicker study on variable speed wind turbines with doubly fed induction generators," IEEE Trans.Energy Convers., vol. 20, pp. 896-905, 12. 2005. [18] J. Maciejowski, Predictive Control with Constraints. Prentice Hall, 2000. [19] Y. Wang and S. Boyd, "Fast Model Predictive Control Using Online Optimization," in Proc. of the 17th World Congress, The International Federation of Automatic Control, Seoul, Korea, pp. 6974-6980, 2008. [20] A. Bemporad, M. Morari and N.L. Ricker, Model Predictive Control Toolbox 3 Users Guide, The Mathworks, Inc., 2008, http://www.mathworks.com/access/helpdesk/help/toolbox/mpc/) [21] M.A. Henson, "Nonlinear model predictive control: current status and future directions," Comput.Chem.Eng., vol. 23, pp. 187-202, 1998. [22] M. Kuure-Kinsey and B.W. Bequette, "Multiple model predictive control: a state estimation based approach," in 2007 American Control Conference, pp. 3739-3744, 2007. [23] G.F. Franklin, J.D. Powell and M.L. Workman, Digital control of dynamic systems. Menlo Park, Calif.: Addison-Wesley, 1998. [24] F. Borrelli and M. Morari, "Offset free model predictive control," in 46th IEEE Conf. on Decision and Control 2007, pp. 1245-1250, 2007. [25] B. Boukhezzar and H. Siguerdidjane, "Comparison between linear and nonlinear control strategies for variable speed wind turbines," Control Eng. Pract., vol. 18, pp. 1357-1368, 2010. [26] Z. Lubosny, Wind Turbine Operation in Electric Power Systems. Berlin Heidelberg New-York: Springer-Verlag, 2003. [27] A. Bemporad, F. Borrelli, and M. Morari, "Min-max control of constrained uncertain discrete-time linear systems," IEEE Transactions on Automatic Control, vol. 48, pp. 1600-6, 2003.

(C.2)

Copyright (c) 2011 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.