See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/279979624

Adaptive Gain Modified Optimal Torque Controller for Wind Turbine Partial Load
Operation

Conference Paper · October 2014

DOI: 10.1115/DSCC2014-5921

CITATIONS READS
4 660

3 authors, including:

Zheren Ma Mohamed L. Shaltout

Citadel Zewail University of Science and Technology
30 PUBLICATIONS   300 CITATIONS    25 PUBLICATIONS   165 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Hydraulics & Flow-Loops & Modeling View project

Modeling and Control of Managed Pressure Drilling View project

All content following this page was uploaded by Zheren Ma on 15 August 2019.

The user has requested enhancement of the downloaded file.

 Proceedings of the ASME 2014 Dynamic Systems and Control Conference
DSCC2014
October 22-24, 2014, San Antonio, TX, USA

DSCC2014-5921

ADAPTIVE GAIN MODIFIED OPTIMAL TORQUE CONTROLLER FOR WIND

TURBINE PARTIAL LOAD OPERATION

Zheren Ma Mohamed L. Shaltout Dongmei Chen*

Dept. of Mechanical Engineering Dept. of Mechanical Engineering Dept. of Mechanical Engineering
University of Texas at Austin University of Texas at Austin University of Texas at Austin
Austin, TX 78712 Austin, TX 78712 Austin, TX 78712
Email: [email protected] Email: [email protected] Email: [email protected]

ABSTRACT tial load below the rated power where the control objective in-
In this paper, an adaptive gain modified optimal torque con- volves energy capture maximization. The power curves under a
troller (AGMOTC) is proposed and evaluated for wind turbine series of wind speeds for the NREL 5MW wind turbine is shown
partial load operation. An internal PI technique is applied for in Fig. 1. The red dash line indicates a fixed optimal tip speed ra-
gain scheduling in order to accelerate the controller response un- tio (TSR) where the wind energy capture is maximized. In order
der volatile wind speed while the adaptive searching technique to improve optimal TSR tracking, the generator torque should be
endows the controller with robust convergence to the optimal op- controlled to accelerate or decelerate the rotor under the variation
erating point under plant uncertainties. The light detection and of wind speed.
ranging (LIDAR) technology is integrated with the AGMOTC to Generally, enhancing the energy capture capability [2, 3]
provide reliable previewed wind speed measurements. Simula- along with mitigating different types of drive-train fatigue load-
tions on the NREL 5MW wind turbine show that the LIDAR- ings [4, 5] have been the main driver for developing new wind
enabled AGMOTC outperforms the baseline controller consid- turbine designs and control algorithms. Specifically, many strate-
ering the wind energy yield. Additionally, the results show the gies for wind turbine control during partial load operation have
impact of the proposed controller on the wind turbine fatigue been developed in recent years. Basically, those strategies are
loads. based on Maximum Power Point Tracking (MPPT) algorithms in
order to yield the maximum energy capture [6, 7]. Some MPPT
techniques rely on the design of a TSR feedback controller for
1. INTRODUCTION direct optimal TSR tracking. The controller is fed with the dif-
Wind power is a clean, renewable and widely distributed re- ference between the measured TSR and its reference optimal
source for electricity generation. According to the new wind en- value. As a result, the wind turbine system is forced to operate
ergy agenda from American Wind Energy Association [1], new at the MPP for maximum energy extraction. Despite the sim-
wind projects with total capacity of 7000 MW will be installed plicity of implementing such controller, it relies heavily on the
per year from 2014 through 2020 and the annual wind energy instantaneous TSR estimation and therefore it is vulnerable to
generation will be doubled in the US by 2020. To make wind en- wind speed measurement errors [8]. Additionally, poor tracking
ergy more cost-competitive, extensive research has been devoted performance could be expected from the direct TSR feedback
to the advanced control technologies of wind turbine. approach due to ignoring the system nonlinearities.
With the advancement in variable speed operation and Other MPPT techniques rely on the standard torque con-
higher capacity generators, more wind turbines operate with par- troller (STC). The generator torque of STC is directly propor-

1 Copyright © 2014 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 03/04/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 4.5
λ = λopt is caused by the control inputs or the varying wind speed [15].
4
Power curves Thus, the HCS algorithm is applicable only when the detected
Vw = 10m/s
wind speed disturbance is very small to the point that its impact
3.5
can be neglected. This undermines the effectiveness of this tech-
3
Vw = 9m/s nique under a more volatile wind speed profile.
Power(MW)

2.5
In this paper, an adaptive gain modified torque controller
Vw = 8m/s
(AGMOTC) is proposed for wind turbine partial load opera-
2
tion. This controller applies an internal PI technique to modify
1.5
Vw = 7m/s
the gain which accelerates the controller response under volatile
wind speed. The adaptive searching technique is integrated to
1 Vw = 6m/s
address the plant parameter uncertainties and thus improve the
0.5 Vw = 5m/s controller robustness. The light detection and ranging (LIDAR)
0
technology is utilized to provide reliable previewed wind speed
4 5 6 7 8 9 10 11 12 13
rotor speed(rpm) measurement. Simulations are conducted based on NREL 5MW
turbine to show the effectiveness of AGMOTC with respect to
FIGURE 1. Power curves under different wind speeds and the optimal energy harvesting as compared to STC. Additionally, the impact
TSR curve of implementing AGMOTC on the wind turbine fatigue loading
is discussed and analyzed.

tional to the square of generator speed, which eliminates the need

of wind speed measurement and thus increases the reliability of 2. WIND TURBINE SYSTEM
the system. For a given wind speed, the STC will adjust the gen- The NREL 5MW wind turbine model is selected to demon-
erator torque to track an optimal reference torque. Due to the de- strate the effectiveness of the proposed controller. The system
ficiencies in determining the accurate STC gain, many controller model parameters are shown in Table 1. The wind turbine aero-
modifications have been proposed in order to improve its abil- dynamics can be simplified as follows
ity to achieve the MPP. For instance, an adaptive gain schedul-
ing torque controller based on STC is proposed in [9, 10]. This
π C p (β , λ ) 2
adaptive searching technique addresses the parameter uncertain- Jr ω̇ = ρair Dr 5 ω − bω − Gr τ (1)
ties and makes the algorithm independent from turbine internal 64 λ3
dynamics. However, in order to eliminate the influence of wind
speed variation, the algorithm needs to evaluate the average per- where ω, Jr , ρair , Dr , β , b, Gr , τ refer to the rotor speed, rotor
formance over a long time horizon that could be as long as 3 inertia, air density, swept area diameter, blade pitch angle, rotor
hours according to [9, 10]. Additionally, ignoring the instanta- damping ratio, gear ratio and generator torque, respectively. λ
neous influence of wind speed, this control law limits its abil- is the tip speed ratio given by λ = ωDr /(2Vw ), where Vw is the
ity to maximize energy harvesting under highly fluctuating wind wind speed. C p denotes the power coefficient that is a nonlinear
profiles. function of blade pitch angle and TSR.
Hill-climb searching (HCS) method, as another MPPT tech- As shown in Fig. 2, the C p surface is obtained using
nique, is applied to wind turbine control in [11–14]. The princi- WT Perf [16], which is a program developed by NREL for
ple of this method is to perturb a control input and observe the wind turbine performance prediction based on blade element mo-
resulting changes in a specific system output parameter. Then, mentum theory. The pentagram on the surface shows the op-
the searching direction of the control input stays the same or goes timal operating point where λ opt = 7.6, β opt = 0 degree and
in the opposite direction based on whether the observed system Copt
p = 0.4868. Generally, the blade pitch angle is fixed at β
opt

parameter is improved or worsen. In most of the available liter- during partial load operation (blade pitch angle control is beyond
atures, the perturbed control inputs are the inverter or converter the scope of this work). The control objective is to quickly and
variables such as input voltage, duty circle etc. while the ob- robustly converge to the optimal operating point under any wind
served system output is the change in the generator output power. speed and plant uncertainties. Additionally, the thrust coefficient
Using electrical power measurement, the control becomes more Ct , shown in Fig. 3, is obtained using [16] due to its important
reliable without the need of mechanical sensors, thus lower cost role in the study of wind turbine fatigue loading. Consequently,
is achieved. However, the fluctuating wind speed significantly the thrust force Ft acting on the rotor is given by [17]
influences the power measurement precision. In order to guaran-
tee the correct searching direction, wind speed detectors are still π
needed to distinguish whether the change of power generation Ft = ρair Dr 2Ct (β , λ )Vw 2 (2)
8

2 Copyright © 2014 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 03/04/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 TABLE 1. System model parameters

Parameter Magnitude 4

3.5

Rotor diameter 125.8801 m 3

2.5
Rotor inertia 3.8759 × 107 kg· m2 2

Ct
1.5
Gear ratio 97 1

Hub height 90 m 0.5

Air density 1.225 kg/m3 −4

−2
0
Cut-in wind speed 3 m/s 2 20
4 15
6 10
Cut-out wind speed 25 m/s 8 5
0
pitch angle(◦ ) TSR
Rated wind speed 11.4 m/s
FIGURE 3. Thrust coefficient versus TSR and blade pitch angle

C (λ ∗ )
where K1 = 64G π
r
ρair Dr 5 p ∗ 3 ; λ ∗ denotes the reference TSR;
λ
K2 and K3 are the internal PI gains. Their selection will be speci-
0.5 fied in the following sections. Additionally, an adaptive approach
0.4
0.3
has been developed to search for the optimal reference TSR and
will be covered in the following subsections.
Cp

0.2
0.1
0

−4 3.1. Stability analysis

−2 First, the stability of the proposed controller is analyzed. By
0
20 combining Eqn. 1 and Eqn. 3, the simplified turbine system dy-
2
15 namics can be expressed as follows
4
10
6
5 
pitch angle(◦ )
8
0 TSR
 Jr ω̇ + bω = Mω 2 with
πρair Dr 5 C p (λ ) C∗p (4)
FIGURE 2.
M = ( λ3 − ) − Gr K2 (λ − λ ∗ )
Power coefficient versus TSR and blade pitch angle 64 λ ∗3

where C∗p refers to the power coefficient at reference TSR

As shown in Fig. 3, Ct monotonically increases with the increase C p (λ ∗ ). According to the definition of TSR, ω can be substi-
of TSR at zero pitch angle. Since the wind turbine is operating at tuted by 2λVw /Dr . Thus, Eqn. 4 can be rewritten as
a fixed pitch angle (β opt = 0) during partial load, the thrust force
will continuously increase with the rise of the thrust coefficient 2Vw b
and the wind speed. λ̇ = Mλ 2 − λ (5)
Jr Dr Jr

Since the value of b/Jr is small enough to be neglected, the sys-

3. CONTROL METHODOLOGY tem dynamics can be further simplified as
An adaptive gain modified torque controller is proposed to
enhance the tracking performance of optimal TSR. An internal 2Vw
PI technique is applied to the difference between the measured λ̇ = Mλ 2 (6)
Jr Dr
TSR and its reference value for gain scheduling. The proposed
control law is as follows
Figure 4 shows the reference power coefficient curve as a
function of TSR and the corresponding cubic function curve of
τ = [K1 + K2 (λ − λ ∗ ) + K3 (λ − λ ∗ )dt]ω 2 TSR. As shown in Fig. 4, when λ < λ ∗ , the solid line is higher
R
(3)

3 Copyright © 2014 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 03/04/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 K2 , expressed in Eqn. 3, is designed as follows. The Jacobian
Cp∗ 3 value of the system is given by
0.5 Cp = λ∗3 λ
Cp = Cp (λ)
2Vw λ 2 ∂ M 2M
0.4 J(λ ) = [ + ] (10)
Jr Dr ∂ λ λ
0.3
Cp

Note that M = 0 at the equilibrium point. Thus, the Jacobian

value around the equilibrium point can be derived as
0.2

2Vw λ ∗ 2 πρair Dr 5 3C p ∗ ∂C p 

0.1
λ
∗ J∗ = − [ ( − ) + Gr K2 ] (11)
Jr Dr 64λ ∗ 3 λ∗ ∂ λ Cp =C∗p
0
0 2 4 6 8 10
TSR By exploiting the properties of the point-wise eigenvalues,
it is possible to apply an extension of the pole-placement ap-
FIGURE 4. Reference power coefficient curve with respect to TSR proach developed for linear systems. Thus, the system dynamics
and the corresponding cubic function of TSR is dominated by the magnitude of J ∗ that can be directly tuned
by changing the controller parameter K2 . A larger K2 can lead
C∗ to faster response and therefore better tracking performance of
than the dash line which indicates C p (λ ) > ∗p3 λ 3 . Contrary, the reference TSR. As a result, the energy capture is improved.
λ
when λ > λ ∗ , the solid line is lower than the dash line which However, a controller with a larger K2 will cause the rise of the
C∗
indicates C p (λ ) < ∗p3 λ 3 . Therefore, the following inequality torque variance and the fatigue loading at the same time. Con-
λ
always holds sequently, a trade-off study should be conducted to achieve an
optimal system performance with respect to energy capture and
C p (λ ) C∗p fatigue loading. An empirical value of K2 for a given wind tur-
( − 3 )(λ − λ ∗ ) < 0 ∀λ 6= λ ∗ (7) bine can be obtained as K2 = 6 × 10−6 D5r ρair /Gr .
λ3 λ∗ The controller parameters K1 and λ ∗ are obtained based on
the wind turbine model. In reality, they may deviate from the
The equilibrium points of the system can be obtained at λ̇ = 0. optimal values due to modeling errors. According to Eqn. 3, K1
So both λ = 0 and λ = λ ∗ meet the requirement. However, λ = 0 is inversely proportional to the cubic of the reference TSR and
indicates that the turbine is stationary. As a result, we only focus directly proportional to the power coefficient at this point. Con-
on the other equilibrium point λ = λ ∗ . For a positive K2 , Eqn. 7 sidering a 15% parameter estimation error for both λ ∗ and C∗p
can be extended as (i.e. λ ∗ = 0.85λ opt , C∗p = 1.15Copt
p ), K1 could have a 87% error
from the optimal value (i.e. K1 = 1.87K1opt ), which will signif-
M(λ − λ ∗ ) < 0 ∀λ 6= λ ∗ (8) icantly undermine the controller efficiency. In order to improve
the robustness, the controller is further enhanced as follows.
An integral gain K3 is used in Eqn. 3 to compensate for the
Define a Lyapunov function V = 12 (λ − λ ∗ )2 . Then its derivative
estimation error in C∗p . Consequently, TSR can track the refer-
is of the form
ence value λ ∗ robustly. The controller can compensate for the
parameter estimation error and track the reference TSR faster
2Vw
V̇ = M(λ − λ ∗ )λ 2 (9) with a larger K3 . However, the overshoot and the torque vari-
Jr Dr ance would be magnified at the same time. Wind speed fluctua-
tion may also influence the optimization searching process. Even
Thus, V̇ = 0 holds only for the equilibrium point and V̇ < 0 though a larger K3 will lead to a better performance for a smooth
holds for all λ 6= λ ∗ . According to LaSalle’s invariance theo- wind speed profile, it is not preferred because a large integral
rem, the globally asymptotical stability of the equilibrium point gain may prolong the system response time. An empirical value
is established. of K3 can be obtained as K3 = 0.02K2 .

3.2. Performance analysis 3.3. Adaptive control approach

After proving the system stability, performance analysis re- From the aforementioned discussion, it is possible to con-
sults are discussed. First, the proportional part of the controller clude that the proposed controller can robustly converge to the

4 Copyright © 2014 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 03/04/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 reference TSR under certain plant uncertainties. In this subsec- 3.4. LIDAR augmented control
tion, the optimal reference TSR is selected for adaptive tuning
A reliable wind speed measurement is significant for the
because it is almost fixed for a given wind turbine and can be
AGMOTC controller. A better performance can be achieved by
roughly estimated by simulations. Therefore, the tuning range
using the average previewed wind speed of few seconds ahead in-
is relatively small. Many optimization methods including HCS,
stead of the instantaneous wind measurement. Fortunately, such
adaptive gain scheduling etc. have been developed to search for
previewed wind measurement is now available due to the recent
the optimal operating point [10, 18, 19]. In this paper, a new
advancement in LIDAR technology. LIDAR augmented control
adaptive method is developed as follows. In every T seconds,
has been discussed in details in [20–23] and proved to be useful
the algorithm checks whether the TSR has converged to its ref-
in improving energy capture in partial load operation. The pre-
erence value based on the following condition
view time of LIDAR system is related to the focal distance and
the wind speed. In this paper, the focal distance of LIDAR is
λavg − λ ∗  < δ
 
(12) chosen to be 80 m and the average wind speed is 8 m/s. As a
result, the preview time is set to be 10 sec ahead of the measure-
where λavg refers to the average TSR during the time period ment taken from the wind turbine hub.
T and δ is a small positive constant to test the convergence of
TSR. If the above condition is not satisfied, the algorithm pro-
ceeds without updating the reference TSR. Otherwise, the adap-
tive method is applied as follows 3.5. The impact of a controller on the wind turbine fa-
tigue loading
 ∗
λ (k + 1) = λ ∗ (k) + ∆λ ∗ (k)

 ∆λ ∗ (k) = η ∗ sign(∆C p,avg )sign(∆λ ∗ (k − 1))
 In addition to maximizing power generation, extension of
α |λavg −λ ∗ |   (13) the wind turbine system service life is one of the major objectives
η = in the wind energy industry. This objective is mainly challenged
 ∆C p,avg 
 δ
∆C p,avg = C p,avg (k) −C p,avg (k − 1) by the different types of fatigue loads acting on the wind turbine
system. In general, a controller will specify how the wind tur-
where α is an influence factor of the step size η. A lower-bound bine system will respond to wind speed variations. The resulting
step size ηmin is selected in order to prevent the adaptation rate control actions will have a significant effect on the loads acting
from slowing down too much. And it is also limited by an upper- on the wind turbine dynamic system. Thus, it is essential to in-
bound ηmax so that the wind turbine would not be significantly vestigate the impact of implementing a new control technique on
influenced by a sudden disturbance or measurement error. C p,avg the wind turbine fatigue loads.
refers to the average power coefficient during the time period
The wind turbine tower and blades are the main components
T . By neglecting the rotor damping in Eqn. 1, C p,avg can be
subjected to critical fatigue loads due to wind speed fluctua-
estimated by
tions [24]. The tower is mainly subjected to two bending mo-
R tf ments. With respect to the rotor plane, the first is an out-of-plane
0.5Jr (ω 2f − ω02 ) + Gr t0 (τω)dt moment called fore-aft moment, while the second is an in-plane
C p,avg = (14)
πρair D2r R tf moment called side-to-side moment. Rotor thrust loads are the
8 t0 Vw3 dt
main reason for the tower fore-aft moments (TFAM), while the
rotor torque fluctuations are the main reason for the tower side-
where t0 and t f denote the start and end time of the period. ω0
to-side moments (TSSM). Similarly, the wind turbine blades are
and ω f are the rotor speed at t0 and t f , respectively.
subjected to two bending moments acting mainly on the root of
The parameters in the proposed adaptive method include the
the blades. The first is an out-of-plane moment called blade flap-
length of time period T , the step size influence factor α, the up-
wise moment (BFM), while the second is an in-plane moment
per and lower bounds of the step size ηmin , ηmax . Since the con-
called blade edgewise moment (BEM). The dominant cause of
dition in Eqn. 12 can be set to ensure an accurate estimation, it is
the blade flapwise moments are the rotor thrust loads, while the
not necessary to select a large value for T as long as it is enough
dominant cause of the blade edgewise moments are the gravita-
to generate reliable average wind speed measurement. For the
tional loads due to self-weight of the blades.
four parameters concerning the step size, their selections should
consider the trade-off between convergence time and searching The effect of implementing the proposed controller on the
accuracy of the algorithm. In this paper, the values of the param- wind turbine tower and blades bending moments will be investi-
eters are selected to be T = 30s, δ = 0.15, α = 50, ηmin = 0.01, gated in the following section. The results will be compared to
ηmax = 0.3, respectively. This adaptive approach will enable a that of the STC which is widely implemented in the commercial
gradual convergence to the optimal TSR. wind turbine systems.

5 Copyright © 2014 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 03/04/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 4. SIMULATION RESULTS 10

(m/s)
Vw
To validate the effectiveness of the proposed controller, sim- 8

ulations are conducted using nonlinear turbine model provided 0 200 400 600
t(s)
800 1000 1200

by NREL FAST Code [25]. The following degrees of freedom 12

(rpm)
10

are enabled: drivetrain mode, generator mode, first and sec-

ω
8
6

ond tower side-to-side modes, first and second tower fore-aft 0 200 400 600 800 1000 1200
t(s)
modes, first and second blade flapwise modes and first blade 4

(MNm)
3
edgewise mode. The controller is implemented based on the

τr
2
1
FAST-Simulink interface. The NREL TurbSim Code [26] is used 0 200 400 600 800 1000 1200
t(s)
to generate a 20-min simulated wind field with a mean speed of

(kNm)
30

8 m/s and 25% turbulence intensity. The turbine fatigue is evalu- 20

τ
10
ated by damage equivalent loads (DEL) which are calculated us- 0 200 400 600 800 1000 1200
t(s)
ing a rainflow-counting algorithm with NREL Mlife Code [27]. 8

The simulation results of both cases, with and without plant un-

TSR
6

certainties, are presented. 4

0 200 400 600 800 1000 1200
t(s)
4

(MW)
Power
3

4.1. Simulation results without plant uncertainties 2

1
STC
The detailed performance comparison between the STC and 0 200 400 600 800
AGMOTC1200
1000
t(s)
AGMOTC controllers without considering plant uncertainties is
shown in Fig. 5, where the top plot provides the average wind FIGURE 5. Performance comparison of the STC and AGMOTC con-
speed profile over the rotor disk. The subsequent subplots com- trollers without considering plant uncertainties
pares the rotor speed ω, rotor torque τr , generator torque τ, TSR,
and generator power , respectively. The fatigue loadings compar-
ison between the STC and AGMOTC controllers without consid- 6
(MNm)
TSSM

ering plant uncertainties is shown in Fig. 6. The subplots com- 2

0
−2
pares TSSM, TFAM, BEM and BFM, respectively. The sum- −4
0 200 400 600 800 1000 1200

marized simulation results are shown in Table. 2. Compared to t(s)

the results of the STC controller, the energy capture is increased

(MNm)

60
TFAM

40
slightly by 0.27% using AGMOTC controller. Additionally, the 20

slight drop in the rotor torque profile due to the implementa- 0 200 400 600 800 1000 1200
t(s)
tion of the AGMOTC controller leads to the decrease of aver- 4
(MNm)

age TSSM by 5.44 % and the TSSM DEL by 19.07%. Imple-

BEM

2
0
menting the AGMOTC leads to the slight rise of the TSR profile, −2

0 200 400 600 800 1000 1200

hence the rotor thrust load. As a result, the average TFAM and t(s)
BFM are increased by 3.26% and 3.16 %, respectively. Addition- 10
(MNm)
BFM

8
ally, the TFAM DEL decreased by 2.23% and BFM increased by 6
4
2.25%. Despite the rise of the average TFAM, the improvement 2 STC
0 200 400 600 800
AGMOTC1200
1000

of the TFAM DEL is basically due to the reduction in the TFAM t(s)

volatility. Since the BEM is dominantly affected by the gravi-

FIGURE 6. Fatigue loading comparison of the STC and AGMOTC
tational loads, it is negligibly affected. Therefore, without con-
controllers without considering plant uncertainties
sidering plant uncertainties, the proposed AGMOTC controller
demonstrates a slight improvement of energy harvesting and an
obvious enhancement with respect to TSSM and TFAM DEL
that a small parameter estimation error will significantly under-
mitigation . However; the average TFAM, the average BFM and
mine the efficiency of the controller.
the BFM DEL are worsened.
The above simulations are conducted based on the assump-
tion that there is no uncertainties in the system model. In real 4.2. Simulation results with plant uncertainties
world operation, however, the unpredictable frictions and esti- Next, the simulation results will be shown by using the same
mation errors in power coefficient surface may lead to inaccurate controller parameters and assuming a 15% estimation error in
prediction of reference TSR and the corresponding power coef- turbine optimal C p and λ . Figure 7 compares the performance of
ficient. From the analysis in the previous section, it was proven STC and AGMOTC controllers under plant uncertainties, while

6 Copyright © 2014 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 03/04/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 TABLE 2. Summarized simulation results without plant uncertainties 10

(m/s)
Vw
8
Evaluation parameters STC AGMOTC 6
0 200 400 600 800 1000 1200
t(s)
Energy capture (×109 J) 2.5615 2.5684

(rpm)
10

ω
8

RMSE of TSR error 0.4152 0.4089 6

0 200 400 600 800 1000 1200

t(s)
TSSM DEL (MNm) 7.7499 6.2717 4

(MNm)
3

τr
2
TFAM DEL (MNm) 35.5376 34.7454 1
0 200 400 600 800 1000 1200
t(s)
BEM DEL (MNm) 6.3589 6.3306 40

(kNm)
30

τ
20

BFM DEL (MNm) 5.2552 5.3736 10

0 200 400 600 800 1000 1200

t(s)
8

TSR
TABLE 3. Summarized simulation results with 15% parameter esti- 6

opt 4
mation error (i.e. λ ∗ = 0.85λ opt , C∗p = 1.15C p ) 0 200 400 600
t(s)
800 1000 1200

(MW)
Power
Evaluation parameters STC AGMOTC 3
2
1
STC
0 200 400 600 800
AGMOTC1200
1000

Energy capture (×109 J) 2.3070 2.5372 t(s)

RMSE of TSR error 1.6652 0.6594

FIGURE 7. Performance comparison of the STC and AGMOTC con-
TSSM DEL (MNm) 18.1801 7.2801 trollers considering plant uncertainties
TFAM DEL (MNm) 38.7682 38.2071
BEM DEL (MNm) 6.3233 6.4392
(MNm)

10
TSSM

BFM DEL (MNm) 4.2249 4.9902 −10

0 200 400 600 800 1000 1200
t(s)
60
(MNm)
TFAM

40

Figure 8 compares the impact on the fatigue loadings. The sum- 20

0
marized simulation results are shown in Table. 3. Under plant 0 200 400 600
t(s)
800 1000 1200

uncertainties, implementing the AGMOTC increases the energy

(MNm)

4
BEM

capture by 9.98% as compared to that of the STC. Additionally, 2

0
the drop in the rotor torque profile leads to the decrease of av- −2

0 200 400 600 800 1000 1200

erage TSSM by 8.34 % and the TSSM DEL by 60%. The rise t(s)
of the TSR profile, hence the rotor thrust load, leads to the in- 10
(MNm)
BFM

8
crease of the average TFAM and BFM by 14.24% and 14.52 %, 6
4

respectively. Additionally, the TFAM DEL decreased by 1.45% 2 STC

0 200 400 600 800
AGMOTC1200
1000

and BFM increased by 18.11%. The TFAM DEL is improved t(s)

due to the reduction of TFAM volatility. Therefore, with con-

FIGURE 8. Fatigue loading comparison of the STC and AGMOTC
sidering plant uncertainties, the proposed AGMOTC controller
controllers considering plant uncertainties
improves energy harvesting, TSSM mitigation, and TFAM DEL
reduction. However; those improvements come at the expense of
the average TFAM, the average BFM, and BFM DEL.
and the energy yield becomes 10% less than that from the STC
with accurate system parameters. With AGMOTC controller,
4.3. Comparing the simulation results of the cases the RMSE of TSR error is only magnified slightly and thus the
with and without plant uncertainties wind energy loss is 1.21% compared to AGMOTC with accurate
The aforementioned results show that the STC is signifi- system parameter estimation initially. Under plant uncertainties,
cantly influenced by the parameter estimation error since the gain the TSSM DEL of STC quickly deteriorates and almost doubles
of torque controller has a 87% deviation from its optimal value. while that of AGMOTC is only slightly influenced. Therefore,
In such condition, the RMSE of TSR error almost quadruples the AGMOTC shows better robustness under plant uncertainties

7 Copyright © 2014 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 03/04/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 over STC with respect to both energy yield enhancement and ture and prolonged gear life,” Proc. ASME 5th Dynamic
fatigue loads due to TSSM alleviation. However, those improve- Systems and Control Conference, pp. 33-39.
ments come at the expense of aggravation of some fatigue loads. [6] Abdullah, M.A., Yatim, A.H.M. and Tan, C.W., 2011, “A
Consequently, a next step is to study the trade-off between en- study of maximum power point tracking algorithms for
ergy harvesting and fatigue loading through controller tuning. wind energy system,” IEEE Conference on Clean Energy
and Technology (CET), pp.321-326.
[7] Abdullah, M.A., Yatim, A.H.M., Tan, C.W. and Saidur, R.,
5. CONCLUSION 2012, “A review of maximum power point tracking al-
In this paper, an adaptive gain modified torque controller gorithms for wind energy systems,” Renewable and Sus-
(AGMOTC) is proposed for wind turbine partial load operation. tainable Energy Reviews,16 (5), pp .3220-3227.
An internal PI technique is applied to the standard torque con- [8] Kazmi, R. and Guo, H.-J., 2010, “Review and critical
troller. Thus, the local eigenvalue of controller around the equi- analysis of the research papers published till date on maxi-
librium operating point is modified and the controller response mum power point tracking in wind energy conversion sys-
is accelerated under varying wind speed. This leads to a bet- tem,” 2010 IEEE Energy Conversion Congress and Ex-
ter tracking performance of reference TSR and a better energy position, pp .4075-4082.
capture performance of the controller. An adaptive technique is [9] Johnson, K.E. and Balas, M.J., Pao, L.Y. and Fingersh,
then applied to search for the optimal reference TSR which en- L.J. 2004, “Methods for increasing region 2 power cap-
ables the controller to gradually converge to the optimal operat- ture on a variable-speed wind turbine,” Journal of solar
ing point under plant certainties. Integrating LIDAR technology energy engineering, 126 (4), pp. 1092-1100.
guarantees smooth and reliable previewed wind speed measure- [10] Johnson, K.E., Pao, L.Y., Balas, M.J. and Fingersh, L.J.,
ments for the proposed controller. Simulation results show that 2006, “Control of variable-speed wind turbines: stan-
the LIDAR-augmented AGMOTC will enable the wind turbine dard and adaptive techniques for maximizing energy cap-
to quickly and robustly converge to its optimal operating point. ture,” IEEE Control Systems, 26 (3), pp. 70-81.
However, the enhancements in energy capture came at the ex- [11] Koutroulis, E. and Kalaitzakis, K., 2006, “Design of a
pense of some fatigue loads aggravation. As a future work, a maximum power tracking system for wind energy con-
trade-off analysis is required to investigate how much sacrifice in version applications,” IEEE Transactions on Industrial
energy capture can alleviate the fatigue loads through controller Electronics, 53(2), pp. 486-494.
tuning. [12] Wang, Q. and Chang,L., 2004, “An intelligent maxi-
mum power extraction algorithm for inverter-based vari-
able speed wind turbine systems”, IEEE Transactions on
Power Electronics, 19 (5), pp. 1242-1249.
ACKNOWLEDGMENT [13] Barakati, S.M., Kazerani, M. and Aplevich, J.D., 2009,
This research was supported by National Science Founda- “Maximum power tracking control for a wind turbine sys-
tion (NSF)-Control Systems Program CAREER Award CMMI- tem including a matrix converter,” IEEE Transactions on
1056020. Energy Conversion, 24 (3), pp. 705-713.
[14] Soetedjo, A.,Lomi, A. and Mulayanto, P., 2011, “Mod-
eling of wind energy system with MPPT control,” Pro-
REFERENCES ceedings of the 2011 International Conference on Electri-
[1] http://www.awea.org/windenergyagenda. cal Engineering and Informatics, pp. 1-6.
[2] Hall, J. F., Mecklenborg, C. A., Chen, D., Pratap, S. [15] Kazmi, S. M. R. , Goto,H. , Guo, H. J. and Ichinokura,
B., 2011, “Wind energy conversion with a variable-ratio O., 2011, “A novel algorithm for fast and efficient speed-
gearbox: design and analysis,” Renewable Energy, 36 sensorless maximum power point tracking in wind energy
(3), pp 1075-1080. conversion systems,” IEEE Transactions on Industrial
[3] Hall, and Chen, D., 2012, “Performance of a 100 kW Electronics, 58 (1), pp. 29-36.
wind turbine with a Variable Ratio Gearbox,” Renewable [16] Buhl, M.L., 2004, “WT Perf User’s Guide,” National
Energy, 44, pp 261-266. Renewable Energy Laboratory, Golden, CO.
[4] Shaltout, M., Hall, J., and Chen, D., 2014, “Optimal con- [17] Bianchi, F.D., De Battista, H., Mantz, R. J., 2006, Wind
trol of a wind turbine with a variable ratio gearbox for turbine control systems: principles, modeling and gain
maximum energy capture and prolonged gear life,” Jour- scheduling design, Springer.
nal of Solar Energy Engineering, 136 (3), pp.031007-1–7. [18] Song, Y.D., Dhinakaran, B., and Bao, X.Y., 2000, “Vari-
[5] Shaltout, M. L., Zhao, N., Hall, J. F., and Chen, D., 2012, able speed control of wind turbines using nonlinear and
“Wind turbine gearbox control for maximum energy cap- adaptive algorithms,” Journal of Wind Engineering and

8 Copyright © 2014 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 03/04/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Industrial Aerodynamics, 85 (3), pp. 293-308.
[19] Freeman, J. and Balas, M., 1999, “An investigation of
variable speed horizontal axis wind turbines using di-
rect model reference adaptive control,” Proceeding of the
18th ASME Wind Energy Symposium, pp. 66-76.
[20] Wang, N., Johnson, K. E., and Wright, A. D., 2013,
“Comparison of strategies for enhancing energy capture
and reducing loads using LIDAR and feedforward con-
trol,” IEEE Transactions on Control Systems Technol-
ogy, 21 (4), pp. 1129-1142.
[21] Schlipf, D., Schlipf, D. J.,and Kühn, M., 2013, “Non-
linear model predictive control of wind turbines using LI-
DAR,” Wind Energy, 16 (7), pp. 1107-1129.
[22] Harris, M., Hand, M. and Wright,A.D., 2005, “LI-
DAR for Turbine Control,” NREL/TP-500-39154, Na-
tional Renewable Energy Laboratory, Golden, CO.
[23] Wang, N., Johnson, K.E., and Wright, A. D., 2012, “FX-
RLS-Based feedforward control for LIDAR-enabled wind
turbine load mitigation,” IEEE Transactions on Control
Systems Technology, 20 (5), pp. 1212-1222.
[24] Burton, T., Jenkins, N., Sharpe, D., and Bossanyi, E.,
2011, Wind energy handbook , John Wiley & Sons.
[25] Jonkman, J.M. and Buhl, M.L., 2005, “FAST
User’s Guide,” National Renewable Energy Laboratory,
Golden, CO.
[26] Jonkman, J.M., 2009, “TurbSim user’s guide: Version
1.50,” National Renewable Energy Laboratory, Golden,
CO.
[27] Hayman, G.J., and Buhl, M.L., 2012, “Mlife User’s
Guide”, National Renewable Energy Laboratory,
Golden, CO.

9 Copyright © 2014 by ASME

DownloadedViewFrom:
publicationhttp://proceedings.asmedigitalcollection.asme.org/
stats on 03/04/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use