Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Optimal Control of Underwater Kite Power Systems

Luı́s Tiago Paiva Fernando A.C.C. Fontes


SYSTEC–ISR SYSTEC–ISR
Faculty of Engineering, University of Porto Faculty of Engineering, University of Porto
School of Engineering, Polytechnic of Porto Porto, Portugal
Porto, Portugal Email: faf@fe.up.pt
Email: ltpaiva@fe.up.pt

Abstract—We address the problem of generating electricity α angle of attack [rad]


through Underwater Kite Power Systems. For this problem, φ azimuthal angle [rad]
we develop an optimal control problem formulation using a φref reference angle [rad]
continuous–time model of the kite to devise the trajectories and ω weight coefficient
controls for the kite that maximize the total energy produced u control vector
in a given time interval. This is an highly nonlinear problem
for which the optimization is challenging. We also develop a x state vector
numerical solution scheme for the optimal control problem based
on direct methods and on adaptive time–mesh refinement. We
report results that show that the problem can be quickly solved I. I NTRODUCTION
with a high level of accuracy when using our adaptive mesh Nowadays renewable energy alternatives for large scale
refinement strategy. The results confirm the values of electrical power production are being pursued by companies and re-
power that can be produced with such device.
search groups all over the world. The need for clean energy
Index Terms—Nonlinear systems, Optimal control, Real–time
optimization, Continuous–time systems, Adaptive algorithms, resources is growing at least as the same rate as the inevitabil-
Time–mesh refinement, Kite power systems, Underwater kites, ity of reducing the use of fossil fuels.
Airborne wind energy. The movement of masses of water, both in oceans and in
rivers, contains an immense energy that in its vast majority is
still unexploited. We discuss the means of extracting electrical
N OMENCLATURE power from an ocean tidal stream or river stream using an
  underwater kite (a tethered wing). The main concept is to
A wing reference area of kite m2 use a controlled kite by unwinding a cable coiled around a
AR wing aspect ration   drum connected to a generator in a fixed position. The water
at tether reel–out acceleration m s−2 current drags the controlled kite while pulling the tether in
cD aerodynamic drag coefficient an horizontal plane, generating electrical power (see Fig. 1).
cL aerodynamic lift coefficient However, since the length of the tether is finite, eventually
E energy produced [Ws] the cable has to be coiled back. Therefore, we consider two
Faer aerodynamic forces [N] phases: the reel–out phase in which power is produced when
Fdrag aerodynamic drag force [N] the cable is unwind and the reel–in phase in which power
Fcent centrifugal force [N] is consumed when coiling back the tether. Nevertheless, the
Fcor Coriolis force [N] whole cycle is done in such a way that the balance – the total
Flift aerodynamic lift force [N] power produced in the two phases – is positive.
Finert inertial forces [N] Most of the kite power systems being developed are based
Fth tether force [N] on the concept of exploiting the increase Crosswind Kite
m mass [kg] Power described by Loyd in 1980 [1]. This feature is supported
P power produced [W] by the fact that the aerodynamic lift is proportional to the
RGL rotation matrix from G to L square of the (apparent) fluid velocity
r tether length [m]  Flift = 1/2ρcL (α)Ava2 . (1)
ρ fluid density kg m−3
s wing span [m] Therefore, the maximum power extraction occurs when the
T tether tension [N]  −1 
kite is moving at high speeds in a plane which is orthogonal
va apparent fluid velocity
 ms to the fluid direction. (The kite behaves much like the tip of
vw fluid velocity m s−1  
the blade in a turbine.) The principle is valid when the fluid is
vt tether reel–out velocity m s−1 either air or water. While in water the velocities attained are
978-1-5386-2064-9/17/$31.00 ©2017 IEEE
Magnus effect.
The main objective of the Underwater Kite Power Systems
(UKPS) is to maximize the total energy produced in a certain
time period and, taking into account the problem dynamics,
we select Optimal Control as the most adequate tool to
address this maximization problem. We consider the optimal
control problem (OCP) with a continuous–time model of an
underwater kite power system. To our knowledge, this is
the first time such problem is solved as an optimal control
problem.
When solving this type of OCP, characterized by being
highly nonlinear, with state constraints, we cannot expect to
obtain an explicit analytical solution and even the numerical
solution can be quite challenging. With this in mind, we have
chosen to solve this problem using direct methods [11] which
are known to be relatively robust. When using direct methods,
we start by selecting a time–mesh which, subsequently, in
our approach, is adaptively refined to achieve a solution with
a desired level of accuracy. The optimal control problem is
discretized in the selected time–mesh and then transcribed into
a nonlinear programming problem, which can be solved using
vw
one of the several commercially available optimization solvers
(IPOPT, WORHP, etc.). The adaptive mesh–refinement (AMR)
strategy used considers, in the optimal control problems con-
Fig. 1. Underwater Kite Power System (UKPS). Top view. text, several refinement criteria in a multi–level scheme [12].
Details of this technique and its application in other nonlinear
systems is reported in [12] and [13].
typically much lower, this is compensated by the fact that the From our simulations, we can state that the problem can
fluid density ρ is more than 800 times higher in water. be successfully solved confirming the possibility of producing
In [2] the dynamics of a kite power system installed in a electrical power with such device. We have attained an average
river is devised considering a kinetic energy storage device. power of 4 kw (with instant peaks of up to 30 kw) when using
In [3] numerical simulations are used to study the dynamic a 2 m2 kite in a water stream of 3 m s−1 . The refinement
motion of tethered undersea kites. In that work, a rigid–winged strategy resulted in higher accuracy levels and yet with lower
kite moving in an ocean current is considered. overall computational time, when compared with the use
We are considering that the power generator is mounted in a of traditional meshes having equidistant–spacing nodes in
fixed position at the point of attachment of the tether. However, continuous–time OCP, and with a priori discretized versions
there is another approach in which the generator is mounted of the OCP. In particular, the time to obtain a solution with
on the underwater kite to harness hydrokinetic energy from a similar level of accuracy was 6% of the time needed using
an ocean current when the kite moves in high-speed, cross– equidistant–spacing meshes. The benefits of using an adaptive
current motions, [4]. time–mesh are particularly more evident in highly nonlinear
In 2004, T.W. Grinsted and M.J. Watchorn filed a patent systems, as is the case of the controlled underwater kite and
which relates to a prime mover, an apparatus and method for other nonholonomic systems. The technique led to the use of
extracting power from moving water such as tidal flows and finer meshes in time intervals where sharp turns of the kite
river currents [5]. occur. The problem was solved with the AMR strategy and
Kite Power Systems can also be related to aerial kites. the results were obtained with a time–mesh with only 14%
Airborne Wind Energy Systems (AWES), mainly using kites of nodes yet producing the optimal solution 16× faster and
or rigid wing systems have been proposed, in the recent years, with the same level of accuracy when compared against an
by a growing number of research groups in Europe and the US equidistant–spacing mesh.
[6]. The single–tethered kite with the generator on the ground, This paper is organized as follows. In section II, we develop
analogous to the one being proposed here for the water, is a model for the UKPS based on the study of the acting
one of the most simple, most researched, and it appears to be forces. In section III, we describe the problem of optimizing
one of the most promising. Some examples of Airborne Wind power production in a given time period using an underwater
Energy Systems that are being developed by start–ups are kite power system and we formulate it as an optimal control
AmpyxPower (Netherlands, [7]), KiteGen (Italy, [8]), Makani problem. In section IV, we provide numerical results obtained
Power – now part of Google (USA, [9]). In Portugal, Omnidea by solving the optimal control problem using direct methods
[10] has been developing an AWES based on exploiting the via an adaptive time–mesh refinement algorithm. In section V,
we summarise the conclusions. y

II. K ITE P OWER S YSTEM M ODEL


Flift
In this section, we develop a state–space dynamic model
vw
of the Underwater Kite Power System (UKPS). We model
the forces acting on the kite in a polar coordinate system Fth Fdrag
va ṗ
positioned at the center of the kite, similarly to what was
done in the case of wind kites [14] and [15]. We consider r
two coordinate systems:
• Global G: An inertial cartesian coordinate system (x, y) T
on the horizontal plane where the origin is at the point φ
of attachment of the tether and x is aligned according to x
the fluid direction vw = (vw , 0) – basis (ex , ey ).
• Local L: A non–inertial polar coordinate system (r, φ) –
Fig. 2. Forces acting on the kite
basis (er , eφ ).
Considering the position
   
x r cos(φ) and the aerodynamic coefficients – cL and cD – were chosen
p= = , (2)
y r sin(φ) considering a thin plate [16].
In the local coordinate system
the rotation matrix from L coordinate system to G is
     
RLG = er eφ (3) r̈ −rφ̇2
  p̈ = + (11)
cos(φ) −sin(φ) rφ̈ L 2ṙφ̇ L
= (4)

sin(φ) cos(φ) −1/m
Finert
and the rotation matrix from G coordinate system to L is
RGL = R−1 
LG = RLG .
where the second term is −1/mFinert with Finert representing
Let us consider the fluid velocity vw and apparent velocity the inertial forces in the local coordinate system
of the fluid (water) va acting on the kite  2 
Finert = m rφ̇ (12)
va = vw − ṗ (5) −2ṙφ̇ L
       2  
v ṙ rφ̇
= RGL w − =m +m
0
. (13)
0 G L rφ̇ L 0 L −2ṙφ̇ L
 


cos(φ) vw − ṙ
= . (6) 
Fcent 
Fcor
−sin(φ) vw − rφ̇ L
We assume that its radial component va,r is strictly positive Now, we can write
and that the kite body is at all times positioned in such a way  

that its longitudinal axis is aligned with the apparent fluid m = Fth + Faer (α) + Finert . (14)
rφ̈
velocity, i.e.−va /va .
We assume that we can control directly the acceleration
of the tether (at ) and the angle of  attack (α) of the kite,
Considering the position of the kite (p) and its mass (m),
therefore we can define the state x = r, φ, ṙ, φ̇ , the control
the total force acting on it, decomposed as in Fig. 2, is
u = (at , α) and the dynamic equation
mp̈ = Fth + Faer (α) . (7) ⎡ ⎤
r
Denoting by T the tension on the tether at the base, we d ⎢ ⎥
φ
⎢ ⎥
have ẋ =
  dt ⎣ ṙ ⎦
Fth = −T er = −T (8) φ̇
0 L ⎡ ⎤

Faer (α) = Flift (α) + Fdrag (α) ⎢ ⎥
⎢ φ̇ ⎥
= 1/2ρAva  (cD (α)va + cL (α)R90 va ) (9) =⎢ at ⎥. (15)
⎣ ⎦
1  aer
where R90 is the 90◦ anticlockwise rotation matrix,
inert
Fφ (α) + Fφ
 2 mr
2 2
va  = (cos(φ) vw − ṙ) + sin(φ) vw + rφ̇ (10)
III. O PTIMAL C ONTROL P ROBLEM B. Equilibrium state
The main objective of the UKPS is to maximize the total We use as initial state for the optimizer an equilibrium state
energy produced in a time frame. Given the dynamic nature of with p̈ = 0, ṗ = 0 and u = 0, that is
the model developed in the previous section, we select Optimal
p = (r, φ) = (r0 , φ∗ ) , (26)
Control as the most adequate tool to address this maximization
problem. where r0 can be any positive value and φ∗ satisfies
We consider the problem of optimizing power production in
a given time period t ∈ [0, tf ] (see e.g.[17], [11] as references φ (α) = 0 .
φ̇ = φ̈ = 0 ⇒ Faer (27)
on optimal control and on the numerical methods to find Since with ṗ = 0, we have va = vw , thus
solution of such problems.)
1/2ρAv
w
2
(cL (α) cos (φ∗ ) − cD (α) sin (φ∗ )) = 0 (28)
A. Maximum Power Production Problem
yielding
Let us consider the Tension of the tether on the base  
cL (α)
T = −Fth = Faer Finert − mat . φ∗ = tan−1 . (29)
r r (α) + r (16) cD (α)
On the one hand, we want to maximize the power produc- Considering a symmetrical kite, cL (0) = 0, and therefore a
tion during one reel–out cycle. The instant power production possible solution is α = 0 and φ∗ = 0.
is given by P (t) = ṙT and the energy in the interval is
 tf
E(tf ) = P (t) dt . (17) IV. N UMERICAL R ESULTS
0 We consider the simulation parameters of the kite system
On the other hand, we need to ensure that φ has a cer- to be as follows
tain periodic behavior. We guarantee that by minimizing its parameter value
distance to a given reference curve (φref ), i.e.
ρ 1000 kg m−3
Minimise (φ(t) − φref (t)) .
2
(18) vw 3 m s−1
m 100 kg
2
Since ṙT and (φ − φref ) differ by several orders of mag- A 2 m2
nitude we introduce a weight coefficient ω such that both AR 2
expressions have roughly the same scale. with the hydrodynamic coefficients for small values of α when
Aiming maximum power production while φ follows the the kite is considered to behave as a thin plate [16], i.e.
reference curve φref , we address the following optimal control
problem 2πα
cL (α) = (30)
 tf 2α
1+
Maximise
2
ṙT − ω (φ − φref ) dt (19) AR
0 c2L (α)
cD (α) = 1.28 sin(α) + . (31)
subject to dynamic constraints 0.7πAR
⎡ ⎤ ⎡ ⎤ Considering t ∈ [0, 3] and the reference curve
r ṙ  
d ⎢ ⎥ ⎢ φ̇ ⎥ 2π k t
ẋ = ⎢φ ⎥ = ⎢
⎢ a

⎥ := f (x, u) (20)
φref = M sin (32)
dt ⎣ ṙ ⎦ ⎣  t
⎦ 3
1
φ̇ φ (α) − 2mṙφ̇
Faer we evaluate 
mr 3
E(3) = ṙT dt (33)
input constraints 0

−amax ≤ at ≤ amax ∀t ∈ [0, tf ] (21) for different values of k and M obtaining the results shown
in Table I for M = π/9.
−αmax ≤ α ≤ αmax ∀t ∈ [0, tf ] , (22)

left end–point constraints TABLE I


 P RODUCTION E STIMATES
x0 = r0 , φ0 , ṙ0 , φ̇0 (23) k E(3)
1 1.1870 × 103
and bounded–state constraints 2 4.6322 × 103
3 6.0061 × 103
rmin ≤r(t) ≤ rmax ∀t ∈ [0, tf ] (24) 4 6.4037 × 104
−φmax ≤φ(t) ≤ φmax ∀t ∈ [0, tf ] . (25) 5 1.2673 × 104
Since the highest production is achieved when k = 4, we
fix  
8πt
φref = /9 sin
π (34)
3
and the problem (P )
 3
2
Maximise ṙT − ω (φ − φref ) dt
0
subject to
⎡ ⎤

⎢ φ̇ ⎥
⎢ ⎥
ẋ = ⎢ at ⎥
⎣  ⎦
1/mr
φ (α) − 2mṙφ̇
Faer
Fig. 4. Azimuthal angle
x(0) = x(3) = (5, 0, 1, 0)

−0.1 ≤at ≤ 0.1 ∀t ∈ [0, 3]


−π/180 ≤α ≤ π/180 ∀t ∈ [0, 3]
5 ≤r ≤ rmax ∀t ∈ [0, 3]
− /5π ≤φ ≤ 2/5π
2 ∀t ∈ [0, 3]
is solved using two time–meshes:
• πML : an adaptive mesh generated by the itaretive re-
finement strategy in [12] considering an error threshold
εmax
x = 5 × 10−3
• πF : an equidistant spacing mesh considering the lowest
Δt of πML
The results can be seen in Figs. 3, 4, 6, 7 and 8. The
trajectory of the underwater kite is shown in Figs. 3 and 4, Fig. 5. Control
while the controls are drawn in Fig. 5. In Fig. 6, we can see
the instant power and the energy produced. In Fig. 7 and 8,
the velocities and forces are shown, respectively, along each According to Table II, the mesh πML has only 14% of
of the local coordinates. the nodes of πF , nevertheless both meshes have maximum
absolute local errors of the same order of magnitude. In fact,
computing the solution using πML is more than 16× faster
when compared against the time needed to get a solution using
πF , causing significant savings in memory and computational
cost.

Fig. 3. Optimal Trajectory (top view)

The efficiency of the optimal control algorithms is reported


in Table II where we can find information about the number
of nodes, the smallest time step, the number of iterations
needed to solve the nonlinear programming (NLP) problem,
the objective functional, the maximum absolute local error of
the trajectory, and the CPU times for solving the OCP problem
and for computing the local error as well. Fig. 6. Power and Energy
energy produced in a given interval. We report results that
show that with an adaptive mesh refinement strategy the
problem can be solved in a faster way and with high level
of accuracy. The problem was solved with the AMR strategy
and the results were obtained with a time–mesh with only
14% of nodes yet producing the optimal solution 16× faster
and with the same level of accuracy when compared against
a mesh with equidistant–spacing nodes.
ACKNOWLEDGEMENTS
We acknowledge the support of
FEDER/COMPETE/NORTE2020/POCI/FCT funds through
grants UID/EEA/00147/2013|UID/IEEA/00147/006933–
Fig. 7. Kite Velocities SYSTEC, NORTE–01–0145–FEDER–000033–Stride, and
PTDC–EEI–AUT–2933–2014|16858 –TOCCATA.
R EFERENCES
[1] Miles L. Loyd. Crosswind kite power. Journal of Energy, 4(3):106–111,
1980.
[2] Matthew R. Douglas and Mario W. Gomes. Dynamics of a River
Kite Power Production System with a Kinetic Energy Storage Device.
In Proceedings of the International Conference of Control, Dynamic
Systems, and Robotics (Ottawa, Canada, 2014), pages 1–6, 2014.
[3] Amirmahdi Ghasemi, David J. Olinger, and Gretar Tryggvason. Compu-
tational simulation of the tethered undersea kites for power generation.
In ASME 2015 International Mechanical Engineering Congress and
Exposition, pages V06BT07A043–V06BT07A043. American Society of
Mechanical Engineers, 2015.
[4] H. Li, D. J. Olinger, and M. A. Demetriou. Control of a tethered
undersea kite energy system using a six degree of freedom model. In
2015 54th IEEE Conference on Decision and Control (CDC), pages
688–693, December 2015.
Fig. 8. Kite Forces [5] Timothy William Grinsted and Michael John Watchorn. Ex-
tracting power from moving water, February 2005. U.S. Clas-
sification 290/42, 290/43, 290/53, 290/54; International Classifica-
V. C ONCLUSIONS tion F03D5/06, F03B17/06; Cooperative Classification F05B2210/16,
F03B17/06, Y02E10/28; European Classification F03B17/06.
We have addressed the problem of generating electricity [6] Uwe Ahrens, Moritz Diehl, and Roland Schmehl, editors. Airborne Wind
through Underwater Kite Power Systems (UKPS). From our Energy. Green Energy and Technology. Springer Berlin Heidelberg,
Berlin, Heidelberg, 2013.
simulations, we can state that the problem can be successfully [7] Ampyx Power: Airborne Wind Energy. http://www.ampyxpower.com.
solved confirming the possibility of producing electrical power [8] KiteGen. http://http://kitegen.com.
with such device. We have attained an average power of 4 kw [9] Damon Vander Lind (Makani Power). Developing a 600 kw airborne
wind turbine. in Airborne Wind Energy Conference 2015, June 2015.
(with instant peaks of up to 30 kw) when using a 2 m2 kite in [10] Ricardo J. M. Penedo, Tiago C. D. Pardal, Pedro M. M. S. Silva,
a water stream of 3 m s−1 . Nuno M. Fernandes, and T. Rei C. Fernandes. High Altitude Wind
Energy from a Hybrid Lighter-than-Air Platform Using the Magnus
We have developed an optimal control formulation, as well Effect. In Uwe Ahrens, Moritz Diehl, and Roland Schmehl, editors,
as a numerical solution scheme, for the problem of generating Airborne Wind Energy, Green Energy and Technology, pages 491–500.
electricity through UKPS. The obtained solution devises the Springer Berlin Heidelberg, 2013.
[11] John T. Betts. Practical methods for optimal control using nonlinear
trajectories and controls for the kite that maximize the total programming. SIAM, 2001.
[12] Luı́s Tiago Paiva and Fernando A.C.C. Fontes. Adaptive time-mesh
refinement in optimal control problems with state constraints. Discrete
TABLE II and Continuous Dynamical Systems, 35(9):4553–4572, April 2015.
E FFICIENCY COMPARISON FOR EACH TIME – MESH [13] Luı́s Tiago Paiva. Numerical Methods in Optimal Control and Model
Predictive Control. Ph.D., Universidade do Porto, 2014.
  [14] M. Canale, L. Fagiano, and M. Milanese. High altitude wind energy
 (j)  CPU time (s)
πj Nj Δtj Ij εx  generation using controlled power kites. IEEE Transactions on Control
∞ Solve εx
Systems Technology, 18(2):279 –293, 2010.
π0 16 1/15 46 1.055 × 100 1.771 0.141 [15] Moritz Diehl. Real-Time Optimization for Large Scale Nonlinear
π1 76 1/75 63 2.354 × 10−1 6.360 0.565 Processes. Ph.D., Univ. Heidelberg, 2001.
π2 160 1/375 135 1.561 × 10−2 25.412 1.199 [16] NASA Glenn Research Center. Aerodynamics of kites.
π3 263 1/1875 129 2.619 × 10−3 38.437 1.985 https://www.grc.nasa.gov/WWW/K-12/airplane/kiteaero.html.
Accessed: 2017-02-01.
πML 263 1/1875 286 2.619 × 10−3 71.980 3.890 [17] Richard B. Vinter. Optimal Control. Springer, 2000.

πF 1876 1/1875 586 1.953E−3 1.186E3 14.600

You might also like