Active Structural Control of A Floating Wind Turbine 2017 Journal of Sound

Journal of Sound and Vibration 410 (2017) 447e472
Contents lists available at ScienceDirect
Journal of Sound and Vibration

journal homepage: www.elsevier.com/locate/jsvi
Active structural control of a oating wind turbine with a

stroke-limited hybrid mass damper
Yaqi Hu*, Erming He
School of Aeronautics, Northwestern Polytechnical University, Xi'an 710072, China
a r t i c l e i n f o a b s t r a c t
Article history: Floating wind turbines are subjected to more severe structural loads than xed-bottom
Received 16 March 2017 wind turbines due to additional degrees of freedom (DOFs) of their oating foundations.
Received in revised form 19 July 2017 It's a promising way of using active structural control method to improve the structural
Accepted 24 August 2017
responses of oating wind turbines. This paper investigates an active vibration control
strategy for a barge-type oating wind turbine by setting a stroke-limited hybrid mass
damper (HMD) in the turbine's nacelle. Firstly, a contact nonlinear modeling method for
Keywords:
the oating wind turbine with clearance between the HMD and the stroke limiters is
Floating wind turbine
Dynamic modeling
presented based on Euler-Lagrange's equations and an active control model of the whole
Active vibration control system is established. The structural parameters are validated for the active control model
Hybrid mass damper (HMD) and an equivalent load coefcient method is presented for identifying the wind and wave
Linear quadratic regulator (LQR) disturbances. Then, a state-feedback linear quadratic regulator (LQR) controller is designed
Contact nonlinear to reduce vibration and loads of the wind turbine, and two optimization methods are
combined to optimize the weighting coefcients when considering the stroke of the HMD
and the active control power consumption as constraints. Finally, the designed controllers
are implemented in high delity simulations under ve typical wind and wave conditions.
The results show that active HMD control strategy is shown to be achievable and the
designed controllers could further reduce more vibration and loads of the wind turbine
under the constraints of stroke limitation and power consumption. V-shaped distribu-
tion of the TMD suppression effect is inconsistent with the Weibull distribution in practical
offshore oating wind farms, and the active HMD control could overcome this short-
coming of the passive TMD.
2017 Published by Elsevier Ltd.
1. Introduction
As a kind of rich and important renewable energy, wind energy has become one of the most promising energy and
attracted people's increasing attention in recent years [1,2]. Currently, most wind power is still generated from land-based
wind turbines which are generally installed in vast and sparsely populated lands. However, in many countries, such as
China, America and others, most residents live in coastal areas where land are valuable and relatively rare while power
demand is huge. Therefore, developing offshore wind energy is a good choice, as it can save more land resources and reduce
* Corresponding author.
E-mail address: ah1985@163.com (Y. Hu).
http://dx.doi.org/10.1016/j.jsv.2017.08.050
0022-460X/ 2017 Published by Elsevier Ltd.
448 Y. Hu, E. He / Journal of Sound and Vibration 410 (2017) 447e472
Nomenclature
kp, kt, kT the equivalent spring stiffness coefcients of the platform, the tower and the TMD
dp, dt, dT the equivalent damping coefcients of the platform, the tower and the TMD
mp, mt, mT the masses of the platform, the tower and the TMD
Rp, Rt, RT the distances from the tower hinge to the mass centers of the platform, the tower and the TMD
kUlim, kDlim the equivalent spring stiffness coefcients of upwind and downwind stroke limiters
cUlim, cDlim the damping coefcients of upwind and downwind stroke limiters
xUlim, xDlim the distances of the neutral static HMD to the upwind and downwind stroke limiters
DxU, DxD the deections of the upwind and downwind stroke limiters
xT the displacement of the TMD relative to the z-axis
xnT the displacement of the TMD relative to the neutral static position in the nacelle coordinate
qp, qt the rotation angles of the platform and the tower relative to the z-axis
fa the driving force of the actuator
fk , fd the stiffness restoring force and the damping force of the HMD
g gravitational acceleration
Qlim the damping force of the stroke limiters
Vlim the potential energy of the stroke limiters
Mwind, Mwave the bending moments caused by the wind and wave loads
Acronyms and abbreviations

LM LevenbergeMarquardt
SSE sum of squared errors
TwrBsMxt side-side tower base bending moment
TwrBsMyt fore-aft tower base bending moment
TTD tower top displacement
PtfmPitch platform pitch angle
PtfmRoll platform roll angle
TTDspFA tower top fore-aft displacement
TTDspSS tower top side-side displacement
GenPwr generator power
HmdPwr HMD driving power
BldPitch1 blade pitch angle
TmdDxn TMD/HMD displacement
TmdVxn TMD/HMD velocity
LQR linear quadratic regulator
OoPDe blade-tip out-of-plane displacement
RootMyc1 blade root out-of-plane moment
PGAc parallel genetic algorithm with constraints
STD standard deviation
RMS root mean square
Ave average
power transmission loss. More important is that offshore wind resources are known to be of higher quality than that on land
[3]. Thus global wind power development has been gradually shifting to the sea. Shallow sea wind power has been developed
in recent years, but they are often criticized for visual and noise pollution [4], and foundations of shallow sea wind turbines
are also relatively huge and with complex structures and in high construction costs [5]. By comparison, with less space
limitations and more strong and steady wind resources, deep sea wind power has great potential to be exploited.
Foundation structures of offshore wind turbines play a key role in the sea wind energy development. According to
different foundation types, offshore wind turbines can be broadly categorized into two types: nearshore xed-bottom wind
turbines and deep sea oating wind turbines [6]. The rst one is installed on xed-bottom foundations, including monopile,
gravity based structure, and suction bucket [5]. Currently, these foundations are the mature construction method, and they
are suitable for installation in shallow sea, but not meet economic feasibility in deep sea with depth more than approximately
60 m. The second type uses oating platforms and mooring lines as the supporting structures, which are economical and
feasible for deeper water up to 900 m [7]. Nonetheless, oating foundations are still at an early development stage, and many
different platforms are in proof-of-concept study and they are being tested with scale model in real or laboratory conditions,
including type of barge, spar, tension-leg (TLP) and semi-submersible [5,8].
Y. Hu, E. He / Journal of Sound and Vibration 410 (2017) 447e472 449
It's a new characteristic and also one of the challenges for oating wind turbines that the platform's tilt motion is induced
by the wind and wave excitation. This tilt motion greatly increases the loads of wind turbine due to the effect of large gravity
and inertial [9]. Large tilt motion has a remarkable inuence on tower top deection, which would give rise to unacceptable
huge loads at tower bottom and blades' roots. Jonkman et al. [10] compared the fatigue damage equivalent loads (DELs) of
three types of oating wind turbine and the same wind turbine model with xed-bottom monopile structure, the oating-to-
xed ratio of the tower base pitching moment DEL is 1.7 for the spar type and 2.5 for the TLP type, and worst of all, the ratio
exceeds 7 for the barge type. This phenomenon can denitely result in higher possibility of structural failures and the cost of
maintenance. Thus, advanced control method is required to reduce the loads and improve the structural responses of oating
wind turbines. In addition, exible foundation characteristics of oating wind turbines determines very low platform motion
frequency, which would results in that the traditional blade pitch control method may give rise to negative damping or
unstable motion [11]. These problems have attracted increasing attention from global academia and industrial community to
study effective control method for reducing the loads of oating wind turbines. Currently, there are two categories of control
methods are proposed to reduce the large loads of oating wind turbines.
The rst method utilizes the blade pitch control strategy to change rotor thrust for load mitigation. Jonkman et al.
[7,10,12,13] performed a large number of researches on three main oating wind turbines using a gain-scheduled propor-
tional-integral approach for designing a baseline collective blade pitch controller. The idea of this approach is that reducing
the controller gain could reduce the blade pitch angle, thus increasing rotor thrust and platform pitch damping. Nielsen et al.
[14] proposed an approach in which a basic pitch controller is augmented by an increment pitch angle controller. The pre-
sented approaches are mainly frequency-based controller design methods. There are several approaches in model based
control for oating wind turbines like the LQ approach in Ref. [15], the H approach in Ref. [16], and the variable power
collective pitch approach in Ref. [17]. In Ref. [18], Namik et al. rst presented an advanced individual blade pitch control for
oating wind turbines, which could suppress more platform motion and reduce more tower loads than the baseline collective
blade pitch controller. Later, many researchers were attracted to study on it [19e21]. Focusing on blades, Staino and Basu [22]
discussed the disadvantages of using active pitch control to alleviate aerodynamic loads on blades and proposed an innovative
dual control strategy combining passive pitch control and active tendons inside the hollow structure of the blades.
Although the blade pitch control methods have certain positive role in improving the response of oating wind turbines,
they still suffer from two shortcomings. Firstly, it reduces the motions and loads of the wind turbine always at the cost of more
blade pitch usage, larger output power uctuation, and increased loads at blades' roots. Secondly, it still results in a relatively
remarkable and unacceptable loads and motions after the implementation of the blade pitch control for some oating wind
turbines, such as the barge type [23,24]. It seems that using blade pitch control methods to reduce the response of oating
wind turbines may be not ideal. Therefore, alternatively, the passive or active structural control methods which have been
successfully applied in civil engineering may be capable.
It is a relatively new eld that using passive or active structural control methods to control the vibrations of oating wind
turbines, and these methods are regarded as a recommended way to extend the service life of oating wind turbines. Murtagh
et al. [25] investigated the passive suppression for a simplied wind turbine with a tuned mass damper (TMD) installed at the
tower top. Colwell et al. [26] proposed the idea of using tuned liquid column damper (TLCD) to reduce the vibration response
of a xed-bottom oating wind turbine with the installation position at tower top. Later, Mensah and Duen ~ as-Osorio [27]
assessed the reliability of this idea. Recently, Ikeda et al. [28] derived a 2-degree-of-freedom (DOF) TMD-blade theoretical
model for suppressing the apwise vibration of the wind turbine blades. Moreover, Li et al. [29] made a test on a nearshore
wind turbine using a ball vibration absorber installed in the nacelle. However, these researches are about vibration reduction
of xed-bottom wind turbines, but the dynamic characteristics of them are very different from the oating wind turbines.
Besides, their investigations are not used the state of the art codes FAST (fatigue, aerodynamics, structures, and turbulence)
[30] for modeling wind turbine models, which may not reect the full coupled nonlinear dynamic characteristics of oating
wind turbines.
Based on the FAST codes, Lackner et al. [23] developed a new simulator FAST-SC for considering structural control design of
wind turbines, which added two TMD DOFs in the kinetic equations of wind turbines for load reduction. Utilizing this code,
they conducted a preliminary study on the application of TMD to the large offshore oating wind turbines [23]. With the
consideration of the oating wind turbine as a lump mass, a TLCD is utilized to suppress its surge motion by Luo et al. [31].
Unfortunately, the pitch motion, which is the most essential motion type, cannot be analyzed by this oversimplied model.
Stewart et al. [32e34] built a 3-DOF dynamic model for three types of oating wind turbine based on Newton's second law of
motion, which installed a TMD in the nacelle and platform for load mitigation. Si established a 5-DOF model for spar-type
oating wind turbine based on the D'Alembert's principle of inertial forces and investigated the vibration and load sup-
pression effect by a TMD installed into the nacelle [35] and platform [36] respectively. Even though the TMD spring and
damping were determined by different optimization methods, all of them did not optimize the TMD mass. He et al. estab-
lished a fully coupled aero-hydro-TMD-structural dynamics model [37,38] and a 3-DOF dynamic model [39] of the barge-type
oating wind turbine based on FAST-SC and Euler-Lagrange's equations respectively, in which the TMD mass was optimized
and the control effect of the TMD installed in nacelle was investigated.
Compared with passive design, recent research shows that active structural control could bring about further load and
motion reduction. The hybrid mass damper (HMD) is a hybrid device consisting of a passive TMD supplemented by an actuator
parallel to the spring and damper. It is a well-known concept in structural control, especially for mitigation of excessive dynamic
response of high-rise buildings subjected to strong wind and earthquake loads, where the HMD has been proven to yield
enhanced damping performance compared with the passive TMD [40]. The concept has been implemented in a number of high-
rise buildings, but its application in wind turbines is rare. To date, the authors are aware of only a few literature on this eld
[41e45]. Lackner and Rotea [41] made a HMD structural control design of a barge-type oating wind turbine by solving an H
loop shaping problem, which is conditionally stable and provides effective damping performance when properly tuned. The
actuator dynamics and control-structure interaction were also considered by them [42]. Namik et al. [43] investigated the
performance and comparison of TMD and HMD for damping of the barge-type oating wind turbine by combined with the FAST
wind turbine model used in Ref. [41] and the actuator model used in Ref. [42]. Si et al. [44] designed a gain scheduling state-
feedback H2/H HMD structural controller for load mitigation of a spar-type oating wind turbine. Li et al. [45] designed a
static output-feedback HMD structural controller for the barge-type oating wind turbine by solving a generalized H problem.
It can be known that the HMD control for oating wind turbines is nearly concentrated in H or its related method. All of these
studies suffer from one critical drawback: they didn't consider the stroke of the HMD. Different from civil engineering struc-
tures, offshore oating wind turbines usually have large displacement or deection due to their soft foundations and the
extreme wind and wave conditions. This will directly result in the stroke of HMD much larger than the displacement of the wind
turbine structures (As seen in Ref. [41], the HMD stroke could achieve unacceptable 40 m). The stroke of the HMD is a critical
constraint for actual applications as the practical limitation on the space available for the HMD.
Motivated by the above mentioned problems, this work will present a nonlinear modelling method for wind turbine
structures with clearances and design a LQR controller for load mitigation of a barge-type oating wind turbine, where a HMD
is installed in the nacelle and the HMD stroke is limited by setting stroke limiters. Modelling, identication, controller design,
weighting coefcient optimization and simulation will be investigated to evaluate the effect of the stroke-limited HMD
structural control method.
The rest sections of this paper are organized as follows. Section 2 introduces the considered barge-type wind turbine,
HMD system and the stroke limiters. A 3-DOF contact nonlinear model caused by clearance between the HMD and the
stroke limiters is established based on Euler-Lagrange's equations. The control subroutine based on Fortran language is
developed for implementing a simulation for HMD structural control of wind turbines directly in FAST-SC rather than using
the traditional FAST-Simulink interface. In Section 3, the passive structural parameters of the wind turbine and the TMD,
which were identied by using the nonlinear least squares LevenbergeMarquardt (LM) algorithm previously, are validated
for the active control model of the wind turbine and the HMD used in this paper. Besides, an equivalent load coefcient
method is presented for identifying the wind and wave disturbances. In Section 4, a state-feedback LQR structural
controller is designed, and two different optimization methods are combined to optimize the weighting coefcients by
selecting the standard deviation of the tower top fore-aft deection as the objective function and considering the stroke of
the HMD and the active control power consumption as constraints. In Section 5, the designed controllers are simulated by
the modied FAST-SC code and the discussions are provided for the simulation results. At last, we draw conclusions and
future works in Section 6.
2. Floating wind turbine model and stroke-limited HMD structural control strategy
The schematic of a barge-type oating wind turbine and a fore-aft HMD with stroke limiters is shown in Fig. 1. The wind
turbine model used for analysis and synthesis is the National Renewable Energy Laboratory (NREL) 5-MW wind turbine [7].
Fig. 1. ITI Energy Barge turbine with a fore-aft stroke-limited HMD in the nacelle.
Table 1
Properties of the 5 MW wind turbine and the barge platform.
Item Value
Rating power 5 MW
Baseline control Variable speed, collective pitch
Cut-in, rated, cut-out wind speed 3 m s1, 11.4 m s1, 25 m s1
Cut-in, rated rotor speed 6.9 rpm, 12.1 rpm
Rotor, hub Diameter 126 m, 3 m
Hub, tower top height 90 m, 87.6 m
Nacelle dimension (L W H) 22 m 6 m 6 m
Platform dimension (L W H) 40 m 40 m 10 m
Rotor nacelle assembly mass 350,000 kg
Tower mass 347,460 kg
Platform mass 5,452,000 kg
Number of mooring lines 8
Line length, diameter 473.3 m, 0.0809 m
Anchor depth, radius 150 m, 423.4 m
Table 2
Adopted parameters of the TMD and the stroke limiters.
mT (kg) kT (N m1) cT (N s m1) klim (N m1) clim (N s m1) xlim (m)

110,700 8815 28,720 500,000 500,000 8
This turbine is widely adopted by many other researchers. The adopted oating support structure is the ITI Energy barge, a
big cubic platform attached to eight mooring lines. The adopted wind turbine and oating platform are also widely used for
proof of concept, loads verication and structural control of offshore wind turbines by many other researchers. Detailed
properties of the wind turbine and the oating platform can be seen in Ref. [7], of which some basic properties are shown in
Table 1.
A HMD is drawn out in Fig. 1 for illustrating the situation that it is installed in the nacelle for reducing wind turbine vi-
bration. Here, kT, cT and mT are the stiffness coefcient, the damping coefcient and the mass of the HMD, respectively. Two
stroke limiters are also drawn out in Fig. 1 for limiting the maximum displacement of the moving HMD mass. klim term is the
stiffness coefcient of the stroke limiters, clim term is the damping coefcient of the stroke limiters, and xlim term is the
distance or clearance between the HMD and the stroke limiters when the HMD is under the neutral static position. The words
U and D before the subscript lim represent upwind and downwind, respectively. Besides, fa is the driving force of the
actuator used for active control. It's known that platform pitch motion is the dominant factor causing load increases for the
barge-type oating wind turbine, so the HMD as shown in Fig. 1 is congured in a way that the mass moves in the fore-aft
direction. In Ref. [39], the authors optimized the parameters of the passive stroke-limited TMD for the barge-type wind
turbine. For this research, the parameters of the passive TMD and the stroke limiters in Table 2 (see also in Ref. [39]) will be
adopted, in which for simplicity, klim, clim and xlim are representative of kDlim kUlim, cDlim cUlim and xDlim xUlim, respectively,
i.e., there are same properties of the downwind stroke limiter and the upwind stroke limiter.
As shown in Fig. 2, a reduced model of the wind turbine consists of masses representing the barge, tower and HMD,
which corresponds to 3 DOF that each one of the barge and tower has a single rotary DOF, and the HMD mass has a single
translational DOF. This model contains enough complexity to model the effects of the HMD on the dynamics of the
structure, and is able to be described by a set of three equations. This model does not consider any other structural
components such as blades, because the main goal for this research is to reduce the tower base loads, and it has been shown
[7,10,41] that blade and other component dynamics have a small effect on tower loads compared to the effects of the 3-
DOFs used in the reduced model.
The barge is connected to a rotary spring and damper representing the hydrostatic restoring force and damping from the
water. The damping from the water is non-linear and complex, including viscous and wave radiation effects, but a linear
approximation can be used for small barge displacements and velocities. The tower also has a rotary spring and damper
representing structural stiffness and damping. The HMD is assumed to slide on a perfectly smooth track along the fore-aft
direction. Fig. 2 is also drawn out for illustrating the motion state of the HMD, no contact with any stroke limiter, contact
with the downwind stroke limiter and contact with the upwind stroke limiter from top to bottom, respectively. DxD and DxU
are representative of the distances after the HMD striking the downwind and upwind stroke limiter, respectively. The stroke
limiter is essentially a large spring and damper that come into contact with the HMD mass at a certain set distance xDlim or
xUlim from the undeected HMD spring position. This makes system highly nonlinear. There need a series of if-else-then
statements used for modeling this contact nonlinear model caused by the clearance between the HMD and the stroke limiter.
These statements are needed to insert in the motion equations such that if the mass is hitting the stroke limiter, an additional
force is applied to the mass and an equal opposite force is applied to the nacelle.
2.1. Dynamic model of a barge-type wind turbine with a fore-aft stroke-limited HMD
Different from the Kane's modelling method used by FAST or FAST-SC [30], the Newton's law of motion used by Refs.
[33e35,42] and the D'Alembert's principle of inertial forces used by [35,36,44], this study will establish the governing
equations for the 3-DOFs dynamic model using the Euler-Lagrange's equations. It will be discovered that our proposed
model based on Euler-Lagrange's equations essentially leads to the same form as that from Kane's, Newton's and
D'Alembert's modelling method, but in this case, the Euler-Lagrange's equations could provide a more simple and clear
modelling process than the Kane's method, and could avoid the complex force vector formulation of the Newton's and
D'Alembert's method.
The Euler-Lagrange's equations of a non-conservative system with n generalized coordinates or DOFs are described
as

d vL vL
Qi i 1; 2; /; n (1)
dt vq_ i vqi
LT V (2)
where T and V are the total kinetic energy and total potential energy of the system, respectively. L is the Lagrange operator. Qi
is the generalized non-potential force.
The total kinetic energy and total potential energy of the Barge-type oating wind turbine can be expressed as Eqs. (3) and
(4), respectively, and the generalized non-potential forces consist of the damping force and the external wind and wave loads
can be expressed as Eq. (5).
1 2 1 2 1
T It q_ t Ip q_ p mT x_2T (3)
2 2 2

1 2 1 x RT sin qt 2 1 2
V kt qt qp kT T kp qp mt gRt cos qt
2 2 cos qt 2 (4)
mp gRp cos qp mT gRT cos qt xT RT sin qt tan qt Vlim
Fig. 2. Simplied model of oating wind turbine with a fore-aft stroke-limited HMD in the nacelle.
8
>
>
>
> Qqt dt q_ t q_ p dT RT x_nT Mwind fa RT Qlim RT
>
>
>
>
>
>
>
> dT RT x_T cos qt RT q_ t xT q_ t sin qt
>
> dt q_ t q_ p dT RT Mwind fa RT Qlim RT
>
>
< cos2 qt

(5)
> Qqp dp q_ p dt q_ t q_ p Mwave
>
>
>
>
> QxT dT x_nT fa Qlim
>
>
>
>
>
>
> x_ cos qt RT q_ t xT q_ t sin qt
>
> dT T fa Qlim
>
: cos2 qt
where
8
Dx2 : xnT > xDlim
1 < Dlim 2D
K

2 : KUlim DxU : xnT < xUlim
Vlim
0 : otherwise
8
< CDlim Dx_D : xnT > xDlim x_nT > 0
Qlim CUlim Dx_U : xnT < xUlim x_nT < 0
:
0 : otherwise
xT RT sin qt
xnT
cos qt
DxD xnT xDlim
DxU xnT xUlim
where q and x are the rotation angle and the displacement of the HMD from the vertical axis z, respectively. k and d are the
equivalent spring stiffness coefcient and the equivalent damping coefcient, respectively. I is the moment of inertia about
the mass center. m is the mass, and R is the distance from the mass center to the tower hinge. The subscripts p, t, and T
represent the platform, the tower, and the TMD, respectively. The superscript n represents the nacelle coordinate. Vlim and Qlim
are the potential energy and the damping force of the stroke limiters, respectively. Mwind and Mwave are the bending moments
caused by the external wind and wave loads acting on the tower base and the platform top, respectively.
Substituting Eqs. (3)e(5) into Eqs. (1) and (2), the contact nonlinear dynamic model of the barge oating wind turbine can
be established as
8
>
> Ip qp kt qt qp kp qp mp gRp sin qp dp q_ p dt q_ t q_.
p Mwave
>
>
>
> It qt kt qt qp kT klim xT RT sin qt xT sin qt RT cos3 qt
>
> .
>
>
< klim xlim xT sin qt RT cos2 qt mt gRt sin qt mT gRT sin qt xT sec2 qt
. (6)
>
> dt q_ t q_ p Mwind RT fa dT clim RT q_ t x_T cos qt xT q_ t sin qt RT cos2 qt
>
> .
>
>
> mT xT kT klim xT RT sin qt cos2 qt klim xlim =cos qt mT g tan qt
>
>
> .
: d c R q_ x_ cos q x q_ sin q cos2 q f
T lim T t T t T t t t a
where
8
< kDlim : xnT > xDlim
klim kUlim : xnT < xUlim
:
0 : otherwise
8
< cDlim : xnT > xDlim x_ nT > 0
clim cUlim : xnT < xUlim x_ nT < 0
:
0 : otherwise

xDlim : xnac
tmd > 0
xlim
xUlim : xnac
tmd < 0
Small angle approximations are appropriate throughout this article because in simulations, none of the platforms exceed
10 degrees of pitch, even in the heaviest wind and wave loadings [32,42], and the angle used for all simulations can also be
shown in Ref. [39] for illustrating this feasible assumption. The simplied equations by using small angle assumption can be
obtained as Eq. (7), in which the klim, clim and xlim are same as that in Eq. (6).
8
>
> Ip qp kt qt qp kp qp mp gRp qp dp q_ p dt q_ t q_ p Mwave
>
>
>
>
< It qt kt qt qp kT klim xT RT qt RT klim xlim RT mt gRt qt
(7)
>
>
> mT gRT qt xT dt q_ t q_ p Mwind RT fa dT clim RT q_ t x_T RT
>
>
>
:
mT xT kT klim xT RT qt klim xlim mT g qt dT clim RT q_ t x_T fa
Eq. (7) can be described as Eq. (8), which is in the form of matrix equation. In Eq. (8), M, D and K are the mass, damping and
stiffness matrix, respectively, and we can see all these three matrixes are symmetric. Fa is the control input matrix repre-
senting the position of the applied active control force. Fd is the external wind and wave input disturbance matrix repre-
senting the acting position of the external wind and wave loads. Const is a constant matrix that is the resulting item caused by
the stroke limiters, i.e., the distance between the neutral nacelle position and the position of undeformed stroke limiters.
MX DX_ KX Fa f a Fd ud Const (8)

where the detailed expanded terms of these matrices are given in the appendix.
From Eq. (7) or (8), we can nd that the HMD control system with stroke limiters of the wind turbine has the charac-
teristics of piecewise stiffness and damping. The stiffness restoring force and the damping force of the HMD mass can be
formulated as Eqs. (9) and (10), respectively, which also has the piecewise characteristics, as shown in Fig. 3.
8 n
n < kT x T : xnT > xDlim
fk xT kT kDlim xT kDlim xDlim : xnT < xUlim
n
(9)
:
0 : otherwise
8 n
n < dT dDlim x_T : xT > xDlim x_nT > 0
n
fd xT dT dUlim x_nT : xnT < xUlim x_nT < 0 (10)

:
dT x_nT : otherwise
2.2. State space representation of the controlled wind turbine system
For the convenience of structural control research, it is necessary to transform the model (8) into the following state-space
representation:
Fig. 3. Stiffness restoring force and damping force characteristic of the HMD mass: (a) Stiffness restoring force; (b) Damping force.

x_ Ax Bf a Bd ud StaConst
(11)
y Cx
where, x XT X _ T is the state vector. y Rtop qt qp qp xT RT qt T is the output measurements representing the
tower top deection in the fore-aft direction, the platform pitch angular and the HMD mass displacement with respect to the
nacelle neutral position from top to bottom, respectively. A is the state matrix. B is the control input matrix. Bd is the wind and
wave disturbance matrix. C DspC VelC is the output state matrix, in which the DspC is the displacement output matrix
and the VelC is the velocity output matrix. The detailed expanded terms of these matrices are given in the appendix.
2.3. Development of the control subroutine in FAST-SC
The FAST simulator [30] has been widely used in scientic research and engineering application. This code is capable of the
fully coupled dynamic modeling and load analysis of both onshore and offshore wind turbines, and system simulation
analysis can be done by selecting different DOFs combinations under various wind and wave conditions. Based on the FAST
code, Lackner et al. developed the FAST-SC code [23] by adding additional structural control module. FAST-SC adds two TMD
DOFs into the dynamic equations of the FAST code, and each TMD mass, spring stiffness, damping, and translation direction
can be dened by users. The FAST-SC code also has the ability of designing active structural control for wind turbines, with
each TMD actuated by external active control force input.
The FAST-SC needs to be compiled as a Simulink S-Function block for structural active control by using the FAST-Simulink
interface, which is a complicated process not easy to get succeed [30]. Different from this traditional implementation of the
FAST-Simulink interface, we directly develop the Fortran subroutine HmdXCntrl () in FAST-SC for implementing the HMD
structural control of wind turbines. We use this modied FAST-SC to estimate the unknown structural parameters, identify
the wind and wave loads and evaluate the control effect of the HMD.
A brief description of the subroutine is given here for clarity. The function of the subroutine is to calculate the HMD control
force and return it to the fully coupled aero-hydro-servo-elastic dynamic equations of wind turbines for the active control
force applying at each time step. The input of the subroutine is the control feedback gain matrix and the state vector, so the
output control force of the subroutine is easy to calculate by multiplying the two input items. The feedback gain matrix can be
get by control algorithm and passed into the modied input le of the FAST-SC automatically by using Matlab. The state vector
can be extracted from FAST-SC at each time step. It is important to note that the extracted state vector in FAST-SC is need to
consistent with that in the state-space model (11). In other words, there are some DOFs in model (11) is not consistent with
that in FAST-SC, so we should make some transformation, such as the DOF qt that is not appeared in FAST-SC, so we need to
transform the FAST-SC's tower top deection in the fore-aft direction xtop into it by the relationship xtop RT sinqt qp .
3. Identication of the wind turbine parameters and the wind-wave loads
In this section, the modied FAST-SC is used to estimate the unknown parameters in the proposed model. In total, there are
8 parameters to be identied, and they are classied into two categories, i.e. U U1 U2 T , where

U1 d p dt kp kt Ip It ;
U2 Mwind Mwave ;
U1 is the structural parameter vector of the wind turbine, and U2 is the load parameter vector of the wind and wave.
3.1. The parameters of the wind turbine
The authors identied the parameters of the wind turbine by using the nonlinear least squares Levenberg-Marquardt (LM)
algorithm in previous research [39]. Here we don't explain the identication process for saving space, and only list the
identication results as Table 3.
Table 3
Parameter identication results of the wind turbine.
Term Value Term Value

dp (N m s rad1) 3.6374 107 kt (N m rad1) 9.7990 109
dt (N m s rad1) 2.1032 107 Ip (kg m2) 1.6945 109
kp (N m rad1) 1.4171 109 It (kg m2) 1.8217 109
Fig. 4. Amplitude-frequency curve of the designed fourth order Butterworth lter.
3.2. Validation of the proposed model and its structural parameters
The parameters listed in Tables 2 and 3 are acquired from the authors' previous research [39], which is based on the passive
structural control system. It's still need to verify that whether these parameters t the proposed active control system. The
modied FAST-SC is used as a baseline tool to make a simulation and the results are compared with the results made by the
proposed model (8) for illustrating the validation of these parameters. The data for validation is obtained by applying a
specic control force fa and recording the tower top fore-aft deection (TTDFA), the platform pitching angular (PtfmPitch) and
the HMD mass stroke (HmdDxn).
According to the proposed model (8), we only enable 3 corresponding DOFs in the modied FAST-SC while disabling all
other DOFs; zero initial conditions, except rotor speed at the rated 12.1 rpm. Both of the proposed model and the FAST-SC
don't consider the external disturbances, i.e. still water and no wind. The control force fa has to pass a low passed lter
which represents the actuator dynamics. Consequently, a fourth order Butterworth lter is designed with passband cutoff
frequency at 1.6 Hz, passband maximum attenuation 3 dB, stopband cutoff frequency at 5 Hz and stopband minimum
attenuation at 60 dB. Fig. 4 shows the amplitude-frequency curve of the designed fourth order Butterworth lter. Fig. 5
shows the input control force in time and frequency domain. Over 95% of the signal power is contained in the fre-
quency range 0 Hze2 Hz, which is consistent with the fact that the main modes of vibration have frequencies below 1 Hz
[23,41].
The input control force of Fig. 5 is applied to the proposed model and the modied FAST-SC respectively to obtain the
time and frequency domain responses of the above mentioned 3 recording results, as shown in Fig. 6. It is seen that these
outputs of the proposed model track well with that of the modied FAST-SC. The slightly difference is the high-frequency
components of the platform pitch angle, which can be attributed to non-linearity in the barge pitch spring and damper in
the modied FAST-SC. In this graph, three spectral peaks are apparent at 0.0364 Hz, 0.0874 Hz, and 0.5404 Hz, corre-
sponding to the vibration modes of the TMD, platform pitch and tower bending respectively. The poles of the proposed
model are shown in Table 4. The table also provides the interpretation for each vibration mode by indicating the uncoupled
degree of freedom giving rise to each mode in the last column. For example, recall that the TMD parameters in Table 2, these
parameters give rise to an undamped frequency 0.045 Hz and damping ratio 46%, which are close to the appropriate values
in the rst row of Table 4.
Fig. 5. HMD force fa used for validation: (a) Time series; (b) Power spectrum density.
Fig. 6. Output comparisons of the proposed model and the modied FAST-SC under HMD control force: (a) Time series; (b) Power spectrum density.
3.3. Five typical combined wind and wave load cases
For the subsequent identication of wind and wave disturbances and control simulation, in this section, we select ve
typical combined wind and wave load cases according to normal sea state data from Ref. [7], as shown in Table 5. Load case
numbers 1 to 5 represent the wind turbine running in different conditions: cut-in wind speed, below rated power, near rated
power, higher than rated power, and cut-out wind speed, respectively. The peak spectral periods are set to 11.8 s corre-
sponding to the peak spectral frequency at 0.08 Hz in all ve load cases. This peak spectral period or frequency is near the
platform pitch mode of the system, so this value represents a serious wave load scenario.
For wind conditions, the wind eld is generated by TurbSim [46], which is a stochastic, full-eld, turbulent-wind simu-
lator. The Kaimal spectra and the power-law wind prole with the power law exponent of 0.14 and the normal turbulence
model with B level turbulence intensity are used according to the IEC61400-3 offshore wind turbine design standard. For each
wind speed, three different wind inputs with different random seeds are generated. For wave conditions, the wave loads can
be set in the modied FAST-SC code. All wave inputs use the JONSWAP spectrum. The signicant wave height is set to 1.7 m for
4 m/s wind and increasingly increases to 5.6 m/s for 24 m/s wind. Also, different random seeds are used for different wave
inputs.
Each simulation uses one specic wind and wave input and lasts 630s, and the output data in rst 30s are not recorded,
waiting for generator torque and blade pitch motion arriving normal operation state. The traditional generator torque and
bladed pitch controller from NREL [7] is used in the form of a dynamic link library for all tests presented in this paper, and all
the results are compared to the results using this baseline controller.
3.4. Parameter identication of the wind and wave load cases
Actually, the wind and wave acting on the oating wind turbine is a complex combined aero-elastic and hydro-elastic
problem, which means there is a coupling between the wind and wave load and the wind turbine's response, i.e. the wind
and wave excitation results in the structural response of the wind turbine, and the resulting structural response can results in
new wind and wave load in reverse, and that cycle repeats. This paper is focused on the implementation of HMD structural
control on the oating wind turbine, and the accurate simulation of the wind and wave load is outside the scope of this paper.
As an alternative, we propose a load coefcient method to identify the selected ve typical wind and wave load cases. For
simplicity, we choose rotor thrust (RotThrust) and wave elevation (WaveElev) as the parameters characterizing the effect of
wind on the blades and wave on the platform, respectively, where RotThrust and WaveElev can be got from the modied
FAST-SC. By introducing a wind load coefcient bwind and a wave load coefcient bwave, the bending moment Mwind in the
proposed model can be expressed as the product of bwind and RotThrust, and the Mwave can be expressed as the product of
bwave and WaveElev. Consequently, the proposed state space model (11) can be replaced as (12), and the unknown parameter
vector U2 can be replaced as U02 bwind bwave .
Table 4
Poles of the proposed model.
Eigenvalue (Hz) Damping ratio (%) Undamped freq. (Hz) Damped freq. (Hz) Uncoupled DOF origin
0.01397 0.0336i 38.4 0.0364 0.0336 TMD system
0.01198 0.0866i 13.7 0.0874 0.0866 Platform pitch
0.00580 0.5404i 1.07 0.5404 0.5404 Tower bending
Table 5
Five typical combined wind and wave load cases.
Load case number 1 2 3 4 5

Hub-height wind speed (m s1) 4 8 12 18 24
Expected signicant wave height (m) 1.7 2.0 2.6 4.0 5.6
Peak spectral period (s) 11.8

x_ Ax Bf a B0d u0d StaConst
(12)
y Cx
T
0 0 0 0 bwind 0 RotThrust
where B0d ; u0d .
0 0 0 bwave 0 0 WaveElev
In this section, specically, we also use the LM algorithm to identify the unknown wind and wave coefcient by mini-
mizing the sum of squared errors (SSE) between the proposed model response and the modied FAST-SC simulation result.
The SSE between outputs from FAST-SC and the proposed model (12) is dened as
1X m X
n h i2
S U02 wj yj ti fj ti ; U02
2
j1 i1
(13)
n h i
1X T
Yti F ti ; U02 W Yti F ti ; U02
2 i1
where m denotes the number of outputs, and n is the data length for one output. yj ti is the jth output of the modied FAST-
SC at time ti, while fj ti ; U02 is the jth output of the proposed model with parameter vector U02 at time ti, Yti and Fti ; U02 are
their vector notations. W represents the diagonal weight matrix for normalization.
The ve typical wind and wave load cases are chosen here for their corresponding load coefcients identication. The
identication procedure has three steps. First, in the modied FAST-SC input le, activate the platform pitch, the rst tower
foreeaft bending and the HMD DOFs, deactivate all the others, input the wind and incident wave load cases as Table 5, and
set the initial motion conditions as zero. Then, run the modied FAST-SC code and obtain the fully coupled nonlinear
simulation result. Second, extract the RotThrust and WaveElev time series as disturbance input in the proposed model (12),
and acquire the simulation result by solving the proposed model. Third, quantify the difference of the responses between
the proposed model and the FAST-SC and estimate all the unknown parameters in U02 . Finally substitute the identied load
coefcients into the proposed model, run the simulation, and compare the results again in order to verify the identied load
coefcients.
The LM algorithm used in this work adopts the embedded lsqnonlin solver in MATLAB. Parameter and cost tolerance is set
as 0.001 when identication accuracy and efciency are both considered. It should be noted that the initial guess of unknown
parameters is important in this nonlinear iterative identication problem, since there are a big amount of parameters to be
estimated compared with the number of outputs, and it will usually lead to a slow convergence process and unsatised result
Fig. 7. Iteration process for load wind and wave coefcient identication.
Fig. 8. Identied bending moment caused by wind and wave.
if the initial values are placed too far away from the best identication [36]. So that it needs to take some trial and error to
produce a rough initial guess for these parameters.
For brevity, here we just show the identication process under load case 2 with one wind and wave seed, and other cases
are similar. Fig. 7 shows the identication iteration process, which can be seen good initial value could make a fast
convergence. Using the identied load coefcients, we can get the identied wind and wave load Mwind and Mwave, as seen in
Fig. 8. The time and frequency domain response comparison between the proposed model and the modied FAST-SC is
illustrated in Fig. 9. It can be seen that the two results coincide well with each other, which shows the effectiveness of the
wind and wave load identication. There are two spectral peaks in the frequency response in Fig. 9(b), and the peak fre-
quencies are happened at 0.0867 Hz and 0.5417 Hz. These two frequencies are close to the vibration modes of the last two row
in Table 4. There doesn't appear the HMD's frequency that listed in the rst row in Table 4, which imply that the HMD's
frequency is same as the platform pitch frequency, i.e. both the vibration frequencies of them are at 0.0867 Hz. This also
explains the resonant vibration absorption mechanism of the TMD or HMD. The identied load coefcients of all ve wind
and wave load cases are listed in Table 6, in which we can see that different load cases have more inuence on the identi-
cation values, but different random seeds have less inuence on them.
4. HMD structural controller design
In this section, we will focus on the design of HMD structural controller for mitigation of vibrations and loads on the wind
turbine. The designed controller must satisfy the cost of active control power and HMD stroke which are the two most critical
constraints in practice. Two design stages will be adopted for this purpose: rst a LQR optimal controller is designed for its
free and convenient denition of objective function of control. It's very easy to introduce the constraints of control power and
Fig. 9. Output comparisons of the proposed model and the modied FAST-SC under wind and wave load: (a) Time series; (b) Power spectrum density.
Table 6
Identied wind and wave load coefcients.
Item Seed Case 1 Case 2 Case 3 Case 4 Case 5

bwind s1 1.304 1.779 2.055 1.910 1.788
105 () s2 1.304 1.791 2.066 1.902 1.805
s3 1.413 1.779 2.071 1.921 1.800
bwave s1 8.900 7.619 5.569 5.162 5.936
103 () s2 9.000 7.850 5.770 4.785 5.054
s3 9.000 8.159 6.488 5.390 5.717
HMD stroke into the LQR's objective function. Then, Based on this objective function with multiple constraints, the weight
parameters are optimized for maximizing the control effect.
4.1. LQR optimal control law design
LQR is one of the optimal control techniques that use state space approach to analyze such a system. Regarding vibration
and load reduction, the tower top fore-aft deection (TTDFA), the platform pitch angular (PtfmPitch), the HMD mass stroke
(HmdXDxn) and the HMD active control force (fa) are included in the control objective function, and the cost performance
index is given by:
Z
1
2
J q1 Rtop qt qp q2 q2p q3 xtmd Rtmd qt 2 ru2 dt (14)
2
0
where q1, q2, q3, r are the weighting coefcients corresponding to the TTDFA, PtfmPitch, HmdXDxn and fa respectively.
In this performance index, TTDFA and PtfmPitch are representative of the index of vibration reduction of the wind turbine,
and TTDFA correlates strongly with the tower base fore-aft bending moment, so it is also a representative of the index of load
reduction. The last two indexes HmdXDxn and fa are representative of the two constraints concerned in this paper. It can be
seen that the weighting coefcients could be allocated rationally for achieving maximum vibration and load reduction effect
and at the same time to satisfy the control power and HMD stroke constraint in practice. Substituting the output of Eq. (12)
into the Eq. (14), the performance index can be expressed as common form:
Z
1
J xT CT Q Cx uT Ru dt (15)
2
0
where Q and R are the weight matrices, and Q is a positive semi-denite matrix and R is a positive denite matrix. The value
of the elements in Q and R is related to its contribution to the performance index J. Since the proposed model (12) is
controllable and observable completely, the LQR control exists a unique solution which makes (15) achieving minimum. Using
(15), the correctional Riccati matrix equation can be obtained:
PA AT P PBR 1 BT P CT Q C 0 (16)
By solving (16), matrix P can be obtained and if it is positive denite, the system will be asymptotically stable and the
optimal feedback gain K and the optimal control force u(t) are obtained as
K R 1 BT P (17)
ut Kxt (18)
Actually, the motion state of the HMD mass includes two stages: before striking the stroke limiters (stage 1) and after
striking the stroke limiters (stage 2). We know the HMD mass mainly operate in stage 1, and stage 2 is a remarkably short time
for large stop stiffness and damping. So we will apply active control forces of Eq. (18) only when the HMD mass operate in
stage 1, i.e. before striking the stroke limiters, and not apply control forces in stage 2.
4.2. Weighting coefcient optimization
For LQR control problem, the weighting matrices Q and R in the performance index have a great inuence on the control
effect, and inappropriate weighting coefcients can lead to the control failure. In order to view a graphical representation of
the function that was being optimized, a surface response method was employed. There are four weighting coefcients in the
performance index, for simplicity, we set q1 q2 and r 1, so only two parameters q1 and q3 need to be optimized.
4.2.1. Surface response plots

The surface response plots, i.e. exhaustive search were created by inputting different combinations of q1 and q3 and
recording the standard deviation reduction percentage of the TTDFA. By inputting a range of spaced q1 and q3, and plotting this
array with the standard deviation reduction percentage of TTDFA on a surface plot, the optimum can be found graphically. The
parameter searching range and interval are chosen when both time consumption and accuracy are considered. This is a thorny
problem in LQR control. Here we take two steps to optimize the weighting coefcients: rst, a wide search range from 100 to
1021 and the 10 time interval, i.e. 100, 101, 102, , 1021 are adopted for reecting the total varying trend of the weighting
coefcient congurations. Then, based on the rough results of the rst step, the small good effect regions are used to make a
detail optimization with an evenly interval.
Here, we also use the proposed model and the modied FAST-SC to optimize the weighting coefcients. The simulation
time for one combination of q1 and q3 is approximately 1 min and 2.7 min for the proposed model and the modied FAST-SC,
respectively, and a full optimization process for each load case may need tens of thousands of parameter combinations to be
resolved, which denitely need a lot of time consumption. For this purpose, we developed the code in Matlab with parallel
running function for simulating different parameter combinations at the same time. We run this optimization on a computer
with 28 cores with all cores opening, which can deals with 28 parameter combinations in one time. This greatly saves the
simulation time, approximately 1/25 time consumption of its serial running.
Fig. 10 shows the nephogram of percentage reduction of s(TTDFA) vs. q1 and q3 under load case 2, compared to the passive
systems. There are a narrow and long belt distribution for conguring the weighting coefcients q1 and q3. The plots between
Fig. 10 (a) and (b) have different results, and the optimization domain in Fig. 10(b) is wider than that in Fig. 10(a), i.e. many
combinations of q1 and q3 are failure on control in the modied FAST-SC for verifying. This implies that the reduced 3-DOF
model is not suitable for optimizing the weighting coefcients for its reduced 3-DOFs, which includes the interaction be-
tween the control force and the 3-DOFs, but can't include the interaction between the control force and some other important
DOFs, such as the blades' pitch DOFs. Fig. 11 is the nephogram of the ratio of control power to output power vs. q1 and q3 under
load case 2, which is also seen the different results of the two models. We should combine the Figs. 10 and 11 to nd ranges
where need detail optimization. The general principle is that the places where values are large in Fig. 10 and small in Fig. 11
need to be selected for further optimization.
Fig. 10. Nephogram of percentage reduction of s(TTDFA) vs. q1 and q2 (load case 2): (a) Modied FAST-SC response; (b) Reduced 3-DOF model response.
Fig. 11. Nephogram of the ratio of control power to output power vs. q1 and q2 (load case 2): (a) Modied FAST-SC response; (b) Reduced 3-DOF model response.
After detail optimization by the modied FAST-SC code, compared to the passive systems, the percentage reductions of
s(TTDFA) and s(TwrBsMyt) with the ratio of control power to output power under all ve load cases are shown in Fig. 12. The
distributions of the percentage reductions of s(TTDFA) and s(TwrBsMyt) are similar, and there is a positive correlation be-
tween them. The distributions of the percentage reductions of s(TTDFA) and s(TwrBsMyt) in load case 1 are different from
other four load cases. In load case 1, the percentage reductions of s(TTDFA) and s(TwrBsMyt) increase with the ratio of control
Fig. 12. Percentage reductions of s(TTDFA) and s(TwrBsMyt) with the ratio of control power to output power: (a) Case 1; (b) Case 2; (c) Case 3; (d) Case 4; (e) Case 5.
power to output power, but the percentage reductions of s(TTDFA) and s(TwrBsMyt) are increase rstly and then decrease
with the increase of the ratio of control power to output power. The maximum percentage reductions of s(TTDFA) and
s(TwrBsMyt) decrease gradually from load case 1 to load case 5. In load case 3, using the ratio of control power to output
power approximately 5% can achieve the maximum percentage reductions of s(TTDFA) and s(TwrBsMyt) at 23% and 26%, and
the rest load cases who is farther away from load case 3, more control power consumption need to be used to achieve the
maximum percentage reductions.
Fig. 13 shows the percentage reductions of s(TTDFA) with the HMD stroke. The stroke is measured using the maximum
downwind (in blue circles) or upwind displacement (in red points) and the 90th percentile of the displacement time series
(in green cross). The maximum downwind or upwind displacement is that the peak amplitude of the HMD strokes, and the
90th percentile value quanties the displacement value that is exceeded 10% of the time, which is a better descriptor of
stroke requirements than the peak amplitude (which occurs once typically) [41]. Note that we take a hard stroke con-
straints 8 m into consideration, which is a clear dividing line in Fig. 13. The probability of the HMD mass hitting the stroke
limiters gradually increases from load case 1 to load case 5 by the same weighting coefcient congurations, especially for
the load case 5, the probability arrives at 100%, i.e. the HMD mass hit the stroke limiters under all weighting coefcient
congurations. The maximum distance over the stroke limiters could arrive at approximately 1.3 m in load case 5 by the
effective weighting coefcient congurations, which shows the stroke limiters work normally and the over-limit distance
is allowable.
Although this exhaustive search could be regarded as a global optimization method and it seems simple, it is still limited
by pre-dened parameter searching range and interval. Besides, it is very computationally expensive, which will take hours or
days long to nish one optimization process. Moreover, there might exists better solution if the parameter interval is not small
enough. Therefore, more intelligent and efcient optimization algorithms are demanded. With this in mind, a more elegant
optimization algorithm is described next.
Fig. 13. Percentage reductions of s(TTDFA) with the HMD stroke: (a) Case 1; (b) Case 2; (c) Case 3; (d) Case 4; (e) Case 5.
Fig. 14. Flowchart of PGAc for the weighting coefcient optimization.
4.2.2. Parallel genetic algorithm with constraints (PGAc)

Genetic algorithm (GA) is an evolution-inspired algorithm that includes mathematical representations of populations,
individual tness, mutation, and even mating to nd an optimum to a problem. This approach starts with randomly generated
population, and individuals with better tness will be selected as the basis of next generation. The improved population will
keep evolving after inheritance, mutation, selection, and crossover procedures until it meets the nal requirement. As a global
optimization method, GA is based on stochastic variables and does not require the derivatives of object function, which brings
the advantages of global evaluation and objective tolerance when compared with other gradient-based local optimization
methods [36]. It usually helps to obtain a better result in optimization problems with non-smooth objective functions, thus is
suitable for the optimization problem in this paper.
The capability of using multiple design variables is introduced so both the weighting coefcients q1 and q3 can be specied
in one individual. The algorithm starts by picking a population of random binary strings for both the q1 and q3 constants.
These designs are kept in pairs throughout the algorithm. The binary strings are simply binary encodings of the integer value
of the corresponding q1 or q3 constant. The length of these strings puts bounds on the variables. The tness function is the
reciprocal of the s(TTDFA). Note that the HMD stroke constraint is considered in the modied FAST-SC, and only the control
power constraint should be considered here. We treat the control power constraint as a penalty function, and the new tness
function is constructed by adding the penalty function.
Once the initial population is chosen, the rst step is to check tness values. The checking of the tness values is where the
majority of the computational time is taken because the modied FAST-SC model must be run for each design. For each
running of the modied FAST-SC code, the running time is approximately 2.7 min, so the serial running for the weighting
coefcient optimization seems infeasible for thousands of code calls. To speed up the computations, the PGAc is developed in
this research based on distribution of workload among multiple cores during the time consuming tness function evaluation
phase, followed by a single central population regeneration. Our model of PGAc exploits the fact that each individual in the
population represents an independent coding of the weighting coefcients and hence its tness function evaluation can be
done independently and concurrently. Next, the algorithm saves the elite and average tness and goes on to the scaling,
crossover, and mutation steps.
The scaling used is called sigma-scaling, which normalizes the tnesses by the standard deviation and removes individuals
which are more than 1 standard deviation below the mean of the population. A roulette wheel uniform crossover is used with
a Pcrossover 0.7. Roulette wheel crossover can be visualized as a weighted roulette wheel. The more t individuals have a
Table 7
Optimization results based on PGAc with the ratio of control power to output power as 10%.
Case number Percentage reduction of Ratio of control power q1 q2 q3

s(TTDFA) (%) to output power (%)
PGA ES
1 15.37 15.13 9.96 3.0238 1011 2.9235 109
2 27.94 27.92 9.28 1.9430 1014 4.8730 1011
3 22.55 22.53 4.66 1.9044 1014 7.9057 1011
4 17.85 17.62 9.27 6.2093 1014 3.6328 1012
5 12.76 10.74 9.93 2.8218 1013 3.7359 1011
Table 8
Design model poles of passive and active system.
Control type Eigenvalue (Hz) Damping ratio (%) Damped freq. (Hz) Uncoupled DOF origin
Optimal passive 0.01198 0.0866i 13.7 0.0866 Platform pitch
0.00580 0.5404i 1.07 0.5404 Tower bending
Active Case 1 0.07349 0.0971i 60.4 0.0971 Platform pitch
0.03932 0.5414i 7.24 0.5414 Tower bending
Case 2 0.11344 0.1906i 51.1 0.1906 Platform pitch
0.66661 0.8467i 61.9 0.8467 Tower bending
0.66826 0.8450i 62.0 0.8450 Tower bending
0.95589 1.0837i 66.1 1.0837 Tower bending
0.35314 0.6172i 49.7 0.6172 Tower bending
larger slice of the wheel, and thus are more likely to be bred in the crossover. Uniform crossover takes random bits from the
strings of both the parents and swaps them. A mutation probability, Pmutation 0.01, per bit is applied, where the algorithm
switches 1% of the bits to the opposite of what it was. The mutation step helps to maintain genetic diversity in the population.
The process is then repeated. This process can be seen in Fig. 14.
The optimization results based on PGA are shown in Table 7. It can be noted that the PGAc gives a better result than the ES
method. The reason is that the ES approach is limited by pre-dened parameter searching range and interval, so there might
exists better solution if the parameter interval is not small enough.
4.3. Comparison of design model poles between passive and active system
Table 8 summarizes the values of the eigenvalues, damping ratios and damped frequencies for the platform pitch and
tower bending DOFs. Note that the design poles both have negative real part between passive and active systems, and the
active system poles are farther away from the imagine axis than the passive system poles, which implies the designed LQR
control can make the system get more degree of stability than the passive system. The mode associated with the platform
Table 9
Percentage reduction of evaluation indices obtained from simulation results.
Index Baseline value Passive TMD Active HMD
Value Rel. Baseline (%) Value Rel. Baseline (%) Rel. Passive (%)
STD TTDspFA (m) 0.326 0.205 37.1 0.159 51.2 22.4
STD TwrBsMyt (kN m) 54,234 34,379 36.6 25,182 53.6 26.8
STD PtfmPitch (deg) 1.851 1.042 43.7 0.749 59.5 28.1
Max TmdDxn (m) e 8.080 e 7.990 e 1.1
Max TmdVxn (m s1) e 2.946 e 5.497 e 86.6
STD TmdDxn (m) e 2.051 e 2.160 e 5.3
STD TmdVxn (m s1) e 0.641 e 1.041 e 62.4
Ave. HmdPwr (kW) e e e 160.4 e e
RMS HmdPwr (kW) e e e 291.3 e e
STD GenPwr (kW) 557.9 496.6 4.9 468.9 7.0 12.9
STD BldPitch1 rate (deg s1) 0.30 0.21 30.0 0.18 40.0 14.3
STD RootMyc1 (kN m) 1769 1523 13.9 1443 18.4 5.3
STD OoPDe (m) 0.975 0.874 10.4 0.841 13.7 3.8
STD TTDspSS (m) 0.046 0.034 26.1 0.032 30.4 5.9
STD TwrBsMxt (kN m) 9018 6571 27.1 6180 31.5 6.0
STD PtfmRoll (deg) 0.314 0.209 33.4 0.197 37.3 5.7
Fig. 15. Percentage of reduction of some main evaluation indices in each wind and wave load case: (a) TTDspFA; (b) TwrBsMyt; (c) PtfmPitch; (d) RootMyc1.
Fig. 16. Stroke and velocity of the TMD/HMD in each load case.
pitch DOF increases damping from 13.7% to 60.4%, 51.1%, 52.3%, 51.5% and 54.8% from load case 1 to 5, respectively, while the
mode associated with the tower bending DOF increases damping from 1.07% to 7.24%, 61.9%, 62.0%, 66.1% and 49.7% from load
case 1 to 5, respectively.
5. Simulation and analysis
In this section, using the modied FAST-SC simulations under realistic loading from the ve typical wind and wave load
cases, we evaluate the performance of the designed structural controller by comparing the results of active HMD and its
corresponding passive TMD and uncontrolled baseline system. All the relevant output parameters are compared with the
values for TMD and baseline simulations by rst taking the average of the evaluation indices across the three simulations for
each load case and then calculating the percentage change in the average values of the evaluation indices compared with the
average of the TMD and baseline values.
Fig. 17. Active control power in each load case.
5.1. Simulation results
Table 9 lists some wind turbine indices obtained from the simulations. Because a wind turbine in practice does not
experience an equal amount of time for each wind speed bin, the evaluation indices are then weighted according to the
Weibull distribution and nally averaged across all the ve wind and wave load cases. The selected parameters of the Weibull
distribution are the same as those in [47]. Twelve evaluation indices are considered: TTDspFA and TTDspSS reect the fore-aft
and side-side deections of the tower top, TwrBsMyt and TwrBsMxt reect the fore-aft and side-side loads of the tower
bottom, PtfmPitch and PtfmRoll reect the fore-aft and side-side rotational motions of the barge platform, TmdDxn and
TmdVxn reect the stroke and velocity of the mass of TMD/HMD, RootMyc1 and OoPDe reect the load of the blade root and
out of plane deection of the blade tip, and HmdPwr is the active control power consumption of the HMD. Besides, we also
consider the GenPwr error and BldPitch1 rate indices, which reect the general quality of generator power production and the
usage extent of the blade pitch actuator, respectively. GenPwr error in the table only accounts for 12e24 m/s mean wind
speed, that is, above the rated wind speed but below the cut out one. In the table, Rel. Ba and Rel. Pa are the corresponding
percentage of reduction relative to the baseline and passive TMD, respectively ( and - denote reduction and increasing,
respectively).
To see more details, we further show in Fig. 15 the exact percentage of reduction of some main evaluation indices for
passive and active control in each wind and wave load case. Fig. 16 to show one of the constraints TMD/HMD stroke in each
Fig. 18. Time responses of some indices in wind and wave load case 2.
load case, besides we also add its velocity plot. Fig. 17 is the other constraint active control power in each load case. Fig. 18 to
show a representative 300 s time responses of some indices for the baseline, passive TMD and active HMD conguration
under the wind and wave load case 2, and their power spectral density (PSD) plots are shown in Fig. 19, with the exception of
the power index.
5.2. Result analysis
From the results presented in Table 9 and Figs. 15e17, we have the following observations and analyses:
a) As is shown the results in Table 9, the largest (absolute) load in the wind turbine is the tower base fore-aft bending
moment (i.e., TwrBsMyt). On the positive side, the addition of structural control systems results in substantial positive im-
pacts on the vibrations, loads and performance of the wind turbine. The primary metrics are the reduction in tower base fore-
aft bending moment, tower top fore-aft deection and platform pitching motion, and the designed active structural con-
trollers are especially effective in reducing these indices. Compared with the passive TMD, the designed active structural
controllers are able to reduce them by 26.8%, 22.4% and 28.1%, respectively. The major reason for these vibration and load
reduction is that the platform pitching motion is reduced by active structural control.
b) It is seen from the results in Table 9, reduction of platform pitch motion also more or less leads to improvements of other
indices that are directly related with it. Less platform pitch vibration implies a smaller variation of the wind speed relative to
the rotor plane, so that the need to adjust the blade pitch angle is reduced, the generator power output is steadier and the
blade root load and blade tip deection are also mitigated. It seemly there is a causal link between the reduction of platform
pitch and other indices. The presented indices related with wind turbine side-side motions and loads are also improved by the
fore-aft structural control, but the reduction effect is approximately 5%, which is relatively much less than those in the fore-aft
direction.
c) From Fig. 15, it is seen that suppression rates of passive TMD present the V-shaped distribution characteristics, i.e.
control rates near the rated wind speed region are apparently lower than that regions near cut-in and cut-out wind speed
case. The reason for this phenomenon has explained in our previous research [39], so we don't explained here. From
Fig. 9(a)e(c), compared with passive TMD suppression, the active HMD control rates show that more vibration and load
reduction can be achieved, especially signicant improving the control effect of regions near the rated wind speed, such as
load case 2, 3 and 4. The control rates of the three most critical indices: TTDspFA, TwrBsMyt and PtfmPitch are improved more
than 20% by the designed HMD control in these three load cases, whereas relatively less control rates (approximately 10%) can
be achieved in the cut-in and cut-out load case, which implies that active HMD control could make up for the shortcoming of
passive TMD (The shortcoming is that the V-shaped distribution of the TMD suppression effect is inconsistent with the
Weibull distribution in practice).
The blue line represent the relative percentage reduction of the active HMD compared with the passive TMD, from
Fig. 9(a)e(c), we can see that the relative reduction is gradually decreasing with the increase of the wind and wave load case
Fig. 19. Frequency results of some indices in wind and wave load case 2.
number, with the exception of the load case 1 for which its less relative reduction is due to the constraint of that no more than
10% control power consumption (we can see the relative reduction of the TwrBsMyt can achieve at 38% when the control
power constraint is 30% from Fig. 12(a)).
From Table 9 and Fig. 9(d), there is limited improvement (less than 10%) on the vibration and load indices that related with
blades, which could explained by the fact that our control objective function is aimed at the vibration reduction of the tower
and platform, and future how to design the objective function included the index related with blades' vibration and loads
worth considering.
d) From Figs. 16, 18 and 19, the maximum strokes (absolute) of active HMD and passive TMD are all within the range of
7.5 me9.5 m for all ve wind and wave cases, which implies the 8 m stroke constraints is effective for limiting the
displacement of the TMD/HMD mass, satisfying the limited space in nacelle in practice. Overall, the maximum and standard
deviation strokes of HMD mass are a little bigger than the passive TMD mass strokes due to the external control force on the
HMD mass that results in a more severe strike into the stroke limiters. But the average stroke of the HMD mass is smaller than
that of the TMD mass, which implies that the designed active controller could enhance the ability of resisting deformation of
the wind turbine system under the wind and wave excitation.
Different from the mass stroke, the average velocities between active HMD and passive TMD are almost same (nearly zero),
whereas the maximum and standard deviation velocities between them are very different. The maximum and standard
deviation velocities of the HMD mass are apparent higher than that of the TMD mass, which implies more structural vibration
energy could be dissipated by the damping from the higher HMD mass velocity, and also explain the mechanism of why active
HMD could be more effective for vibration reduction than passive TMD. Combined with the plots of stroke and velocity, we
can see that our designed controller could magnify the velocity of HMD mass but couldn't increase its stroke due to the pre-
set stroke constraints of 8 m, which is different from active HMD control used in civil engineering structures where they
don't need to consider the constraint of the HMD mass for their enough stroke space supplying.
e) From Figs. 17 and 18, the average output power of wind turbine running below the rated wind speed (i.e. load case 1 and
2) is far less than the rated output power 5 MW, and it will reach the rated output power when wind turbine run above the
rated wind speed (i.e. load case 3, 4 and 5). So wind turbine should be run above the rated wind speed as much as possible for
capturing maximum output power. We also can see the active HMD control power consumptions in each load case in Fig. 17.
There only need a small portion of output power (less than 10%) used for active control consumption, which implies the other
pre-set active control power constraint of the ratio of active control power to output power within 10% is effective, and also
demonstrating that the designed active HMD controller is an economical and possibly practical design.
f) From Fig. 18, it is seen that the use of passive TMD and active HMD structural control does improve the structural
response of the wind turbine, and active HMD could achieve more load and vibration reduction than passive TMD, and this is
most apparent in the platform pitch angle. This shows that the designed active HMD controller has some reliability. Fig. 19
shows some large peaks are concentrated between 0.05 and 0.15 Hz in the baseline case. The frequencies of these peaks
are close to 0.08 Hz, which is the peak spectral frequency. The passive TMD and active HMD clearly reduce these peaks, and
active HMD could also achieve more peaks reduction than passive TMD.
6. Conclusions
This paper presents an investigation into the use of active structural HMD control for a barge type oating wind turbine
with two critical constraints in practical application: one is stroke limitations and the other is active control power con-
sumption. The major contributions and conclusions of this study are:
1) A contact nonlinear modelling method for the oating wind turbine with clearances between the HMD and the stroke
limiters is presented based on Euler-Lagrange's equations, and the established nonlinear model includes the constraint of
HMD stroke. The identied passive structural parameters of the wind turbine and TMD by previous researches based on
nonlinear LM algorithm are validated for the established active controlled model of oating wind turbine.
2) An equivalent load coefcient method is presented for identifying the wind and wave disturbances. The identied load
coefcients of all ve typical wind and wave load cases are also obtained by the nonlinear LM algorithm. The different load
cases have more inuence on the identication values, but different random seeds have less inuence on them.
3) The active control subroutine in Fortran language is developed for implementing a simulation for HMD structural control
of wind turbines directly in the famous high delity nonlinear simulator FAST-SC, rather than using the traditional FAST-
Simulink interface for its complicated compilation process not easy to get succeed.
4) A state-feedback LQR structural controller is designed for load reduction of the wind turbine, and two optimization
methods: surface response plots and parallel genetic algorithm with constrains are adopted to optimize the weighting
coefcients. The optimal weighting coefcients are obtained for all of the ve load cases by the fully coupled DOFs wind
turbine model from the modied FAST-SC code, whereas the reduced 3-DOF model is not suitable for optimizing the
weighting coefcients for it can't include the interaction between the control force and some other important DOFs, such
as the blades' pitch DOFs.
5) Compared with the baseline system, the optimized passive TMD results in tower fore-aft load reductions of 48.5%, 39.8%,
32.6%, 39.3% and 47.9% from case 1 to case 5, respectively, while for the active HMD control, load reductions of 57.5%, 60.7%,
50.4%, 53.1% and 54.6% are achievable, satisfying the pre-set stroke constraints of 8 m and the control power constraint of
the consumption-to-output ratio of the power within 10%. V-shaped distribution of the TMD suppression effect is
inconsistent with the Weibull distribution in practical offshore wind farms, and the active HMD control could make up for
this shortcoming of the passive TMD. Active HMD control is shown to be achievable under the constraints of stroke and
active power.
6) The designed HMD-LQR controllers could increase the velocity of the HMD mass but couldn't increase its stroke due to the
pre-set stroke constraints, which implies more structural vibration energy could be dissipated by the damping from the
higher HMD mass velocity, and also explain the mechanism of why active HMD could be more effective for vibration
reduction than passive TMD. The HMD stroke constraint is a new characteristic of offshore oating wind turbines, which is
different from the active HMD control used in civil engineering structures where they don't need to consider the constraint
of the HMD mass for their enough stroke space supplying.
Starting from this work, future work possibly includes extending the presented active HMD control with the two critical
constraints to the barge platform or other types of offshore oating wind turbines such as TLP, spar-buoy. Also future work
needs to be considered the observer-based LQR control design approach, which is easier to be implemented in practice.
Acknowledgements
This research was funded by the National Natural Science Foundation of China with grant number 51675426 and the
Specialized Research Fund for the Doctoral Program of Higher Education of China with grant number 20116102110002.
The authors also want to thank Dr. Lackner in University of Massachusetts Amherst for his open source code FAST-SC and
Dr. Jonkman from National Renewable Energy Laboratory for his generous help and support with the specications of Barge
model. Besides, the authors wish to thank the reviewers for their valuable comments and suggestions.
Appendix
The detailed expanded terms of these matrices in Eq. (8) are as follows:
2 3
Ip
M4 It 5
mT
2 3
dp dt dt 0
D 4 dt dt dT clim Rtmd dT clim RT 5
2
0 dT clim Rtmd dT clim

2 3
kp mp gRp kt kt 0
K 4 kt mt gRt kt kT klim Rtmd mT gRT kT klim RT mT g 5
2
0 kT klim RT mT g kT klim

8 9 2 3 8 9
< 0 = 0 1 < 0 =
4 5 Mwind
Fa RT ; Fd 1 0 ; ud ; Const Klim xlim RT
: ; Mwave : ;
0 0 0 Klim xlim
The detailed expanded terms of these matrices in Eq. (11) are as follows:

0 I
A
M1 K M1 D
2 3
kp mp gRp kt kt
6 0 7
6 Ip Ip 7
6 7
6 7
1 6 kt mt gRt kt kT klim R2T mT gRT kT klim RT mT g 7
M K 6 7
6 It It It 7
6 7
6 7
4 kT klim RT mT g kT klim 5
0
mT mT
2 3
dp dt dt
6 I 0 7
6 p Ip 7
6 7
6 7
6 dt dt dT clim R2T dT clim RT 7
M1 D 6 7
6 It It It 7
6 7
6 7
4 dT clim RT dT clim 5
0
mT mT
T
0
B 0 0 0 0 RItT 1
M1 Fa mT
2 3T
60 0 0 0 1 07
0 6 Itw 7
Bd 1 6 7
M Fd 40 0 0 1 0 05
Ip
T
klim xlim RT klim xlim
StaConst 0 0 0 0
It mT
2 3
Rtop Rtop 0 0 0 0
C DspC VelC 4 1 0 0 0 0 05
0 Rtmd 1 0 0 0
References
[1] D.Y.C. Leung, Y. Yang, Wind energy development and its environmental impact: a review, Renew. Sust. Energy Rev. 16 (1) (2012) 1031e1039.
[2] A. Chehouri, R. Younes, et al., Review of performance optimization techniques applied to wind turbines, Appl. Energy 142 (2015) 361e388.
[3] B. Snyder, M.J. Kaiser, Ecological and economic cost-benet analysis of offshore wind energy, Renew. Energ 34 (6) (2009) 1567e1578.
[4] J.K. Kaldellis, M. Kapsali, Shifting towards offshore wind energy-recent activity and future development, Energ Policy 53 (2013) 136e148.
[5] C. Perez-Collazo, D. Greaves, G. Iglesias, A review of combined wave and offshore wind energy, Renew. Sust. Energy Rev. 42 (2015) 141e153.
[6] X.W. Li, H.J. Gao, Load mitigation for a oating wind turbine via generalized H structural control, IEEE Trans. Ind. Electron. IEEE Trans. Ind. Electron
63 (1) (2016) 332e342.
[7] J.M. Jonkman, Dynamics Modeling and Loads Analysis of an Offshore Floating Wind Turbine (PhD thesis), University of Colorado, Boulder, USA, 2007.
[8] J.M. Jonkman, D. Matha, Dynamics of offshore oating wind turbinesddanalysis of three concepts, Wind Energy 14 (4) (2011) 557e569.
[9] S. Buttereld, W. Musial, et al., Engineering Challenges for Floating Offshore Wind Turbines (Report number: NREL/CP-500-38776), National
Renewable Energy Laboratory, Springeld, USA, 2007.
[10] J. Jonkman, D. Matha, A Quantitative Comparison of the Responses of Three Floating Platform Concepts (Report number: NREL/CP-500e46726),
National Renewable Energy Laboratory, Golden, USA, 2010.
[11] T.J. Larsen, T.D. Hanson, A method to avoid negative damped low frequent tower vibrations for a oating, pitch controlled wind turbine, J. Phys. Conf.
Ser. 75 (1) (2007) 012073.
[12] J.M. Jonkman, Inuence of control on the pitch damping of a oating wind turbine, 6th AIAA Aerospace Sciences Meeting and Exhibit 7-10 January
2008, (Reno, Nevada).
[13] D. Matha, Modelling and Loads & Stability Analysis of a Floating Offshore Tension Leg Platform Wind Turbine, M.S. thesis, National Renewable Energy
Labs National Wind Turbine Center, Stuttgart, Germany, 2009.
[14] F.G. Nielsen, B. Skaare, et al., Method for damping tower vibrations in a wind turbine installation. United States Patent: 8186949-5-29, 2012.
[15] S. Christiansen, T. Bak, T. Knudsen, Damping wind and wave loads on a oating wind turbine, Energies 6 (8) (2013) 4097e4116.
[16] T. Bakka, H.R. Karimi, S. Christiansen, Linear parameter-varying modelling and control of an offshore wind turbine with constrained information, IET
Control Theory A 8 (1) (2014) 22e29.
[17] M.A. Lackner, An investigation of variable power collective pitch control for load mitigation of oating offshore wind turbines, Wind Energy 16 (4)
(2013) 519e528.
[18] H. Namik, K. Stol, Individual blade pitch control of oating offshore wind turbines, Wind Energy 13 (1) (2010) 74e85.
[19] H. Namik, K. Stol, Individual blade pitch control of a spar-buoy oating wind turbine, IEEE T Contr Syst. T 22 (1) (2014) 214e223.
[20] S. Li, F. Yang, Load optimization control of large deep-water oating wind turbines, J. Mech. 2 (3) (2014) 169e175.
[21] K.T. Magar, M.J. Balas, S. Frost, Direct adaptive control for individual blade pitch control of wind turbines for load reduction, J. Intel. Mat. Syst. Str. 26
(12) (2015) 1564e1572.
[22] A. Staino, B. Basu, Emerging trends in vibration control of wind turbines: a focus on a dual control strategy, Philos. T R. Soc. A 373 (2035) (2015)
20140069.
[23] M.A. Lackner, M. Rotea, Passive structural control of offshore wind turbines, Wind Energy 14 (3) (2011) 373e388.
[24] V.N. Dinh, B. Basu, S. Nagarajaiah, Semi-active control of vibrations of spar type oating offshore wind turbines, Smart Struct. Syst. 18 (4) (2016)
683e705.
[25] P.J. Murtagh, A. Ghosh, et al., Passive control of wind turbine vibrations including blade/tower interaction and rotationally sampled turbulence, Wind
Energy 11 (4) (2014) 305e317.
[26] S. Colwell, B. Basu, Tuned liquid column dampers in offshore wind turbines for structural control, Eng. Struct. 31 (2) (2009) 358e368.
[27] A. Mensah, L. Duen ~ as-Osorio, Reliability analysis of wind turbines equipped with tuned liquid column dampers (TLCD), in: Structures Congress,
Chicago, IL, ASCE, Reston, VA, 29e31 March 2012, pp. 1190e1200.
[28] T. Ikeda, Y. Harata, et al., Vibration suppression of wind turbine blades using tuned mass dampers, in: ASME 2014 International Design Engineering
Technical Conferences and Computers and Information in Engineering Conference, Buffalo, NY, ASME, New York, 17e20 August 2014.
[29] J. Li, Z. Zhang, J. Chen, Experimental study on vibration control of offshore wind turbines using a ball vibration absorber, Energy Power Eng. 4 (3)
(2012) 153e157.
[30] J. Jonkman, M.L. Buhl Jr., Fast User's Guide (Report number: NREL/EL-500e38230), National Renewable Energy Laboratory, Golden, USA, 2005.
[31] N. Luo, L. Pacheco, et al., Smart structural control strategies for offshore wind power generation with oating wind turbines, in: International con-
ference on renewable energies and power quality, Santiago de Compostela, European Association for the Development of Renewable Energy, Envi-
ronment and Power Quality (EA4EPQ), Santiago de Compostela, 28e30 March 2012.
[32] G.M. Stewart, Load Reduction of Floating Wind Turbines Using Tuned Mass Dampers, MSc Thesis, University of Massachusetts, Amherst, MA, 2012.
[33] G.M. Stewart, M.A. Lackner, Offshore wind turbine load reduction employing optimal passive tuned mass damping systems, IEEE T Contr Syst. T 21 (4)
(2013) 1090e1104.
[34] G.M. Stewart, M.A. Lackner, Determining optimal tuned mass damper parameters for offshore wind turbines using a genetic algorithm, in: 50th AIAA
Aerospace Sciences Meeting Including the New Horizons Forum and Aerospace Exposition, Nashville, TN, AIAA, Reston, VA, 9e12 January 2012. AIAA
2012-0376.
[35] Y.L. Si, H.R. Karimi, H.J. Gao, Modeling and parameter analysis of the OC3-Hywind oating wind turbine with a tuned mass damper in nacelle, J. Appl.
Math. 2013 (2013) 10. Article ID 679071.
[36] Y.L. Si, H.R. Karimi, H.J. Gao, Modeling and optimization of a passive structural control design for a spar-type oating wind turbine, Eng. Struct. 69
(2014) 168e182.
[37] E.M. He, Y.Q. Hu, Y. Zhang, Structural vibration control of offshore oating wind turbine based on TMD, J. Northwest Polytech Univ. 32 (1) (2014)
55e61.
[38] E.M. He, Y.Q. Hu, et al., Vibration and load suppression of offshore oating wind turbine, in: The 1st International Conference on Advanced Materials,
Structures and Mechanical Engineering, Incheon, South Korea, TTP, Pfafkon, 3e4 May 2014, pp. 891e896.
[39] E.M. He, Y.Q. Hu, Y. Zhang, Optimization design of tuned mass damper for vibration suppression of a barge-type offshore oating wind turbine, Proc.
Inst. Mech. Eng. M. J. Eng. Marit. Environ. 231 (1) (2016), 1475090216642466.
[40] F. Ricciardelli, A.D. Pizzimenti, M. Mattei, Passive and active mass damper control of the response of tall buildings to wind gustiness, Eng. Struct. 25
(2003) 1199e1209.
[41] M.A. Lackner, M.A. Rotea, Structural control of oating wind turbines, Mechatronics 21 (4) (2011) 704e719.
[42] G.M. Stewart, M.A. Lackner, The effect of actuator dynamics on active structural control of offshore wind turbines, Eng. Struct. 33 (5) (2011) 1807e1816.
[43] H. Namik, M. Rotea, M. Lackner, Active structural control with actuator dynamics on a oating wind turbine, in: 51st AIAA Aerospace Sciences Meeting
Including the New Horizons Forum and Aerospace Exposition, 7e10 January 2013 (Grapevine, Texas).
[44] Y.L. Si, H.R. Karimi, Gain scheduling H2/H structural control of a oating wind turbine, in: Proc. 2014 IFAC World Congress, 2014.
[45] X.W. Li, H.J. Gao, Load mitigation for a oating wind turbine via generalized H structural control, IEEE T Ind. Electron 63 (1) (2016) 332e343.
[46] B.J. Jonkman, Turbsim User's Guide: Version 1.50 (Report number: NREL/TP-500e46198), National Renewable Energy Laboratory, Golden, USA, 2009.
[47] H. Namik, Individual Blade Pitch and Disturbance Accommodating Control of Floating Offshore Wind Turbines, Ph.D dissertation, Univ. Auckland, New
Zealand, 2012.

Active Structural Control of A Floating Wind Turbine 2017 Journal of Sound

Încărcat de

Informații document

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Active Structural Control of A Floating Wind Turbine 2017 Journal of Sound

Încărcat de

Drepturi de autor:

Formate disponibile

Journal of Sound and Vibration 410 (2017) 447e472

Contents lists available at ScienceDirect

Journal of Sound and Vibration

Active structural control of a oating wind turbine with a

Acronyms and abbreviations

mT (kg) kT (N m1) cT (N s m1) klim (N m1) clim (N s m1) xlim (m)

DxD xnT xDlim

DxU xnT xUlim

MX DX_ KX Fa f a Fd ud Const (8)

fd xT dT dUlim x_nT : xnT < xUlim x_nT < 0 (10)

2.2. State space representation of the controlled wind turbine system

2.3. Development of the control subroutine in FAST-SC

3. Identication of the wind turbine parameters and the wind-wave loads

3.1. The parameters of the wind turbine

Term Value Term Value

Fig. 4. Amplitude-frequency curve of the designed fourth order Butterworth lter.

3.2. Validation of the proposed model and its structural parameters

3.3. Five typical combined wind and wave load cases

3.4. Parameter identication of the wind and wave load cases

Load case number 1 2 3 4 5

Fig. 8. Identied bending moment caused by wind and wave.

4. HMD structural controller design

Item Seed Case 1 Case 2 Case 3 Case 4 Case 5

4.1. LQR optimal control law design

4.2. Weighting coefcient optimization

4.2.1. Surface response plots

Fig. 14. Flowchart of PGAc for the weighting coefcient optimization.

4.2.2. Parallel genetic algorithm with constraints (PGAc)

Case number Percentage reduction of Ratio of control power q1 q2 q3

Index Baseline value Passive TMD Active HMD

5. Simulation and analysis

Fig. 17. Active control power in each load case.

5.1. Simulation results

5.2. Result analysis

0 dT clim Rtmd dT clim

0 kT klim RT mT g kT klim

S-ar putea să vă placă și