MaynoothSlides JH AF

Control of the PTO system of OWCs:
feedback vs model predictive control
João C. C. Henriques and A. F. O. Falcão
2017 Maynooth University Wave Energy Workshop / Control of the PTO system of OWCs: feedback vs model predictive control 1/31
Outline of the presentation
PART 1
Generator feedback control: Mutriku test case

Power take-off system: biradial vs Wells turbine
Mathematical model
Generator feedback control law: computation and sensitivity analysis
PART 2
Model predictive latching control: Spar-buoy OWC test case

Power take-off system: biradial turbine equipped with an HSSV
Changes to mathematical model
A discrete control algorithm based on Pontryagin’s Maximum Principle
A new continuous control method based on the Discontinuous Galerkin
Method
Major conclusions
PART1 - Mutriku power plant test case
Hydrodynamic model
Hydrodynamic model developed by Wanan Sheng

High-order method, 2192 variables, 300 frequencies
Installed Mutriku power take-off
Wells turbine with biplane rotor without guide vanes
Rotor diameter: 0.75 m. Generator rated power: 18.5 kW
Biradial turbine to be installed at Mutriku within OPERA H2020 Project
Biradial turbine with fixed guide vanes (Kymaner/IST patent WO/2011/102746)
Rotor diameter: 0.50 m. Generator rated power: 30.0 kW

To be installed at Mutriku in mid-April 2017 and in the Oceantec spar-buoy,
deployed at BIMEP site, in Sept. 2017 (OPERA H2020 Project)
Turbine dimensionless curves
The performance characteristics of a turbine in dimensionless form
p 1.4
Ψ= Π
ρin Ω2 d 2 1.2
η
1.0 Φ
ṁturb 0.8
Φ=
ρin Ω d 3 0.6
Pturb 0.4
[− ]
Π=
ρin Ω3 d 5 0.2
0.0
Π
η= 0.2
ΨΦ 0.4
0.6
0.8
8 6 4 2 0 2 4 6 8
d - rotor diameter Ψ [− ]
For Re > 106 and Ma < 0.3 the dimensionless curves Φ, Π and η are independent
of Re and Ma
⇒ ṁturb (p, Ω, ρin ) = ρin Ω d 3 Φ(Ψ), Pturb (p, Ω, ρin ) = ρin Ω3 d 5 Π(Ψ)
Mathematical model
OWC motion is modelled as a rigid piston (LOWC λwaves )
(m2 + A∞
22 ) ẍ2 = −ρw gS2 x2 −R +F −p S p ∗
| {z } | {z22} | {zd2} | atm{z 2 }
buoyancy radiation diffraction air chamber
Air chamber pressure (see [1, 2])
V̇c (γ−1)/γ ṁturb

ṗ ∗ = −γ (p ∗ + 1) − γ (p ∗ + 1)
Vc ρVc
Turbine/generator set dynamics
I Ω̇∗ = (Tturb − Tgen ) Ω−1 ρin Ω2 d 5 Π(Ψ) − Tgen Ω−1

max = max
Dimensionless variables
p − patm
p∗ = and Ω∗ = Ω/Ωmax
patm
O [x2 ] ≈ O [p ∗ ] ≈ O [Ω∗ ] ≈ 1 ⇒ Similar orders of magnitude for errors
Generator feedback control law
If the turbine operates at the best efficiency point ηmax
Pturb ≈ ρatm d 5 Π( Ψ(ηmax ) ) Ω3 = const Ω3

| {z }
const
the generator should follow the turbine in time-average
Feedback control law
Pgen = min a Ωb , Pgen

rated

!
rated
b−1
Pgen
Tgen = min aΩ ,
Ω
The control law is clipped at generator rated power
Constants “a” and “b” are functions of the WEC hydrodynamics and turbine
geometry but not of sea-state
Usually 2.4 < b < 3.6
To compute “a” and “b” we use a constant rotational speed model
Ω̇∗ = 0 ⇒ very large inertia
for the Mutriku wave climate (14 sea states)
Further details can be found in [1, 2]
Biradial turbine Wells turbine

50 50
Pgen = aΩb [W] Pgen = aΩb [W]
sea states sea states
40 40
30 30
P̄t urb [kW]

P̄t urb [kW]
20 20
10 10
0 0
1000 2000 3000 4000 5000 6000 7000 1000 2000 3000 4000 5000 6000 7000
Ω [rpm] Ω [rpm]
0.8 0.8
Π Π ×100
η η
0.6 0.6
Φ Φ
0.4 0.4
0.2 0.2
0.0 0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 0.00 0.05 0.10 0.15 0.20 0.25 0.30
Biradial turbine power output sensitivity to “a” and “b”
4.0
sea state 3 sea state 7 6.7
0.8 7.5 2k
0.9 5 1k W
3.8 4 k kW 8.3 W
1.0 W 0k
4k W
3.6 W
0.9 1.04 k 7.5 8.30 k

b
3.4 0.7 0.854 kW W 6.71 kW W

0 5 k kW 5 2k
0.4 0.56.65 kW 3.57 4.36 5.1.593 kW
3.2
0.2 0.37 kW kW
6 W 2
2.00 .78 k W k W kWW
7 k kW kW kW
W
3.0
sea state 11 Annual average
21. 3.9
3.8 07 1k
kW 4.3 W
2k
W
3.6
21. 4.3
b
3.4 07 2k
19.1 kW 3.1 3.50 3.91 W
17.1 2 kW 0 kW kW kW
11.313.3 15.2 8 k 1.87 2.22.69
3.2 6 k 0 kW4 kWW
5.54 7.48 9.4W 1.06 1.47 kW 8 kWkW
kW kW2 kW kW kW
3.0 0.1 0.2 0.3 0.5 1.0 2.0 4.0 0.1 0.2 0.3 0.5 1.0 2.0 4.0
×10
4 ×10
4
a a
Wells turbine power output sensitivity to “a” and “b”
3.6
sea state 3 0 sea state 7
0 000..1.003
3.5
0 . 0.40.3.259.2k12W6kkWkWWkW
4 2
0.61 0.54 k8WkWkWW
k 9.280.2765433.21023..3.37044369258
3.4 kW kWkWkkWkkWWW
3.3 9.20
kW
b
0.61 8.23
3.2 0.54 kW 7 kW
kW 6.3 .27 kW
3.1
0.
0.42 48 kW 5.33 0 kW
kW 4.36 kW
kW
3.0
sea state 11 Annual average
3.5
4.23 1 0 0.22
11186846240...6.789 kW 33..32724.28.4k9.01550.k6.16kk17 kkWkW
3.4 123456503928176 k 8 kWWWWW W
kkkWWWW
3.3 4.23
b
kW
3.2 18.1 3.78
6 kW 3.34 kW
16.2 2. kW
3.1 14.3
3 kW W
5k 2.45 89 kW
kW
3.0 1 2 3 4 5 6 7 1 2 3 4 5 6 7
×10
5 ×10
5
a a
Comparison time-averaged turbine output power and rotational speed
Biradial turbine Wells turbine
20
13 14 13 14
.51
kW
3.0 25.97 kW 12 3.0 12
W
W W
Gen. law clipping
11 62
3
9
5
1
.2
.0
.9 4kkk
8 11
W
23.24 kW 80
10
2
4
62
3.9
.7
.6
.5
.3
k
Turbine power
2.5 2.5
8.3
.80
6k
15
W
15 10 10 17.66 kW
.0 17.78 kW 13
6.8
2.0 4 2.0 .9
12
H s [m]
H s [m]
kW
4.6
9 4 9
.0
5k
kW
8
5k
kW
W
W
8 12.31 kW 8
1.5 1.5
7 9.58 kW 7 10.22 kW
6 6
2 3 5 2 3 5
1 4 4
0.
.3 6.51 kW
93
1.0 1 9 1.0 1
kW
kW
4.12 kW
2.79 kW
0.5 0.5
6 8 10 12 14 16 6 8 10 12 14 16
Tp [s] Tp [s]
14 14
237
13 13
d /s
479
ra
462 rad/s
rad
998 rad/s
7
3.0 12 3.0 12
68
rad
417 rad/s
/s
894 rad/s
/s
/s
32 372 rad/s 11 58
2 raddd/s/s
ra
9ra 11
Rotational speed
7 66
1
2
3
4 776
ra
2.5 d/ 2.5 791 rad/s
s
282 rad
/s 10 10
2.0 2.0 583 rad/s
H s [m]
9
H s [m] 9
19
37
1
6
ra
ra
8 8
d/
d/
1.5 1.5
s
s
10
16
7 7
1r
8r
6 6
ad
ad
2 3 5 2 3 5
4 4
/s
/s
1.0 15 1.0 1
6
ra
d/ 146 rad/s 272 rad/s
s
0.5 0.5
6 8 10 12 14 16 6 8 10 12 14 16
Tp [s] Tp [s]
PART 2 - Optimal control of the turbine High-Speed Stop Valve (Latching)
HSSV
rotor
The HSSV valve of a biradial turbine with 0.5 m rotor diameter HSSV movie
Latching will be also tested at Mutriku in June-July 2017

(OPERA H2020 Project)
Latching control and air compressibility
OWCs vs oscillating body WECs
Latching on OWCs not so effective as in the case of an oscillating-body

WECs
air compressibility has a spring effect
closing the HSSV does not stop the relative motion between the OWC
and buoy
Decreases the forces resulting from latching in comparison with solid-body

breaking
p AHSSV p AOWC since AHSSV AOWC
No impact forces - air compressibility spring effect
Removes the constraint of latching having to coincide with an instant of zero

relative velocity between the floater and the OWC
Changes to the mathematical model of Mutriku
Spar-buoy OWC - two body system constrained to oscillate in heave
Buoy: (m1 + A∞ ∞
11 ) ẍ1 + A12 ẍ2 = −ρw gS1 x1 − R11 − R12 + Fd1 +patm S2 p
∗
OWC: A∞ ∞
−R21 − R22 +Fd2 −patm S2 p ∗
21 ẍ1 + (m2 + A22 ) ẍ2 = −ρw gS2 x2 |
| {z } {z } | {z } | {z }
buoyancy radiation diffraction air chamber
HSSV simulation requires a new definition of Ψ

p ∗ pat
Ψ̃ = u Ψ = u
ρin Ω2 d 2
where u ∈ {0,1} is the discrete control (close/open)
System of equations written as a 1st-order ODE
ẋ = f(t,x,u) (1)
x are called the states
Optimal control and the Pontriagyn Maximum Principle
Optimal control
Maximization of a performance index based on the control u of the HSSV

ZT
max L (x,u) dt
0
Pturb (t,x,u)
Example: L (x,u) =
ρatm Ω3max d 5
Using the Pontriagyn Maximum Principle (PMP) the optimal problem is

recast as
 
ZT Z Tf
L (x, u) dt − λT (ẋ − f(t,x,u)) dt 
 
max
| {z } | {z }
 
0 0
performance index constrained to system dynamics
λ 0 s are called adjoint variables
Solution of the Pontriagyn Maximum Principle
The PMP shows that:

along the optimal control path (x,u) the Hamiltonian function H
H (t,x, u, λ) = λT f (t,x,u) + L (x, u) ,
is maximum for the optimal input u subjected to:

ẋ = f(t,x,u)
⇒ forward solution
x(0) = x0

λ̇ = −∇x f(t,x,u)T λ − ∇x L(x, u)T
⇒ backward solution
λ(Tf ) = 0
for t ∈ [ 0,Tf ]
non-causal control - Fd (t) required to compute f for t ∈ [ 0,Tf ]
For complete details see the two classical books of Luenberger [5] and
Bryson & Ho [6]
Receding Horizon control in a real-time framework
Hardware-in-the-loop tests at Tecnalia test rig
See [3, 4] for further details
Hs = 2 m Hs = 4 m
0.35
∆TRH = 24 s
0.30 ∆TRH = 16 s
∆TRH = 10 s
0.25
NL (No latching)
0.20
L∗c [−]
0.15
0.10
0.05
0.00
Te [s]
ε∗c = L∗c − L∗c |NL / L∗c |NL [%]
120
100
80

60
40
20
0
8 9 10 11 12 13 14 15 16 8 9 10 11 12 13 14 15 16
Te [s] Te [s]
3.5
3.0 ∆ TRH = 24 s
2.5
∆ TRH = 10 s
[− ]
2.0
No latching
Pturb
1.5
∗
1.0
0.5
0.0
0.8
0.6
0.4
ηt [− ]
0.2
0.0
-0.2
-0.4
Time [s]
0.2
p∗ [− ]
0.0
-0.2
-0.4
open Time [s]
u [− ]
close
Time [s]
Lessons learned from the Tecnalia hardware-in-the-loop tests
To improve stability, all state variables must have the same order of magnitude
The algorithm sometimes opens/closes the HSSV intermittently during short

periods (less than 1.5 seconds)
The discrete nature of the numerical solution of the PMP problem implies a
2nd-order accuracy
A new challenge in the HSSV control
The sub-optimal problem of open/closing the HSSV only at the end of

each (non-infinitesimal) time step, ∆t = 0.1 ∼ 0.2 s
No solution found in the searched bibliography!
We start the quest for a new solution method...
Discontinuous Galerkin Method
The computational domain is discretized in small time-elements
where
State, adjoint and control
P variables approximated by Legendre polynomials,
pj (t) such that x = j pj (t)x̃j
The numerical solution is continuous within the time elements and

allowed to be discontinuous across element boundaries
Solution of the sub-optimal control problem - compute u that maximizes the

Hamiltonian H in a integral sense
Z te+1
max H(t,x, u, λ) dt
u ∈ {0,1} te
In the DG Finite Element Method, the original problem
ẋ − f(t, x, u) = 0
is replaced by a weak formulation

Z
dxh
vh − f dt + vh (te ) [xh (te+ ) − x(te− )] = 0
dt
Ie
| {z } | {z }
state equations weakly enforced BC
vh is the so-called test function in the FEM framework
Applying an affine transformation from t ∈ Ie to τ ∈ [−1,1]

Z1 Z
dx̂h ∆t 1
v̂h dτ + v̂h (−1) x̂h (−1+ ) = v̂h fˆ dτ + v̂h (−1) x̂(−1− ) (2)
−1 dτ 2 −1
we write (2) as linear system A x̂h = b(t,x,u) in each element
To improve the convergence rate, a regularization term was introduced

Z
1 T Pturb (t,x,u)

2
L(x,u) = + ε (1 − u) dt
T 0 ρatm Ω3max d 5
ε is a small constant
0.0120 0.0120
0.0115 0.0115
0.0110 0.0110
0.0105 0.0105
0.0100 0.0100
∆t = 0.001 s ∆t = 0.001 s
0.0095 ∆t = 0.005 s 0.0095 ∆t = 0.005 s
0.0090 ∆t = 0.010 s 0.0090 ∆t = 0.010 s
0.0085
∆t = 0.050 s 0.0085
∆t = 0.050 s
Power without optimal control ∆t = 0.100 s Power without optimal control ∆t = 0.100 s
0.0080 0.0080
∆t = 0.200 s ∆t = 0.200 s
0.0075 0.0075
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
iteration, n iteration, n
open
u [− ]
close
0.4 ε = 0.0001
ε = 0.05
0.2 no control
p ∗ [− ]
0.0
0.2
0.4
t ime, t [s]
Major conclusions
Receding horizon optimal latching algorithm can be used to improve the OWC
spar-buoy capture width
Receding horizon time interval between 10 to 24 s
The discrete algorithm sometimes opens/closes the HSSV intermittently during

short periods (less than 2 seconds)
The proposed Discontinuous-Galerkin method is an efficient alternative to the

well know Pseudo-Spectral Methods
The DG solves the open/closing intermittency problem of the HSSV valve
Probably the greatest advantage of the OWC technology

Simple control of the available power to PTO system by using a HSSV
HSSV can be used for latching and control of the available power to the
turbine and generator
Acknowledgements
The work was funded by the Portuguese Foundation for Science and Technology
(FCT) through IDMEC, under LAETA Pest-OE/EME/LA0022, by WAVEBUOY
project, Wave-powered oceanographic buoy for long term
deployment, PTDC/MAR-TEC/0914/2014, and also by European Union’s
Horizon 2020 research and innovation programme under grant agreement No
654444 (OPERA Project).
The first author was supported by FCT researcher grant No. IF/01457/2014.
References
[1] A. F. O. Falcão, J. C. C. Henriques, Oscillating-water-column wave energy converters and air

turbines: A review, Renewable Energy 85 (2016) 1391 – 1424.
doi:10.1016/j.renene.2015.07.086.
[2] J. C. C. Henriques, J. C. C. Portillo, L. M. C. Gato, R. P. F. Gomes, D. N. Ferreira, A. F. O.

Falcão, Design of oscillating-water-column wave energy converters with an application to
self-powered sensor buoys, Energy 112 (2016) 852 – 867.
doi:10.1016/j.energy.2016.06.054.
[3] J. C. C. Henriques, L. M. C. Gato, A. F. O. Falcão, E. Robles, F.-X. Faÿ, Latching control of a

floating oscillating-water-column wave energy converter, Renewable Energy 90 (2016) 229 – 241.
doi:10.1016/j.renene.2015.12.065.
[4] J. C. C. Henriques, L. M. C. Gato, J. M. Lemos, R. P. F. Gomes, A. F. O. Falcão, Peak-power

control of a grid-integrated oscillating water column wave energy converter, Energy 109 (2016)
378 – 390.
doi:10.1016/j.energy.2016.04.098.
[5] D. G. Luenberger, Introduction to dynamic systems : theory, models, and applications, J. Wiley
& Sons, New York, Chichester, Brisbane, 1979.
[6] A. E. Bryson, Y.-C. Ho, Applied optimal control: optimization, estimation, and control, Blaisdell
Publishing Company, Waltham, 1969.

MaynoothSlides JH AF

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

MaynoothSlides JH AF

Uploaded by

Copyright:

Available Formats

Control of the PTO system of OWCs:

feedback vs model predictive control

João C. C. Henriques and A. F. O. Falcão

Generator feedback control: Mutriku test case

Model predictive latching control: Spar-buoy OWC test case

Hydrodynamic model developed by Wanan Sheng

Wells turbine with biplane rotor without guide vanes

Rotor diameter: 0.75 m. Generator rated power: 18.5 kW

Biradial turbine with fixed guide vanes (Kymaner/IST patent WO/2011/102746)

Rotor diameter: 0.50 m. Generator rated power: 30.0 kW

The performance characteristics of a turbine in dimensionless form

OWC motion is modelled as a rigid piston (LOWC  λwaves )

Air chamber pressure (see [1, 2])

V̇c (γ−1)/γ ṁturb

Turbine/generator set dynamics

I Ω̇∗ = (Tturb − Tgen ) Ω−1 ρin Ω2 d 5 Π(Ψ) − Tgen Ω−1

Generator feedback control law

If the turbine operates at the best efficiency point ηmax

Pturb ≈ ρatm d 5 Π( Ψ(ηmax ) ) Ω3 = const Ω3

the generator should follow the turbine in time-average

Feedback control law

Pgen = min a Ωb , Pgen

The control law is clipped at generator rated power

Usually 2.4 < b < 3.6

To compute “a” and “b” we use a constant rotational speed model

Ω̇∗ = 0 ⇒ very large inertia

for the Mutriku wave climate (14 sea states)

Further details can be found in [1, 2]

Biradial turbine Wells turbine

P̄t urb [kW]

0.9 1.04 k 7.5 8.30 k

3.4 0.7 0.854 kW W 6.71 kW W

Biradial turbine Wells turbine

Latching will be also tested at Mutriku in June-July 2017

OWCs vs oscillating body WECs

Latching on OWCs not so effective as in the case of an oscillating-body

air compressibility has a spring effect

Decreases the forces resulting from latching in comparison with solid-body

p AHSSV  p AOWC since AHSSV  AOWC

No impact forces - air compressibility spring effect

Removes the constraint of latching having to coincide with an instant of zero

Spar-buoy OWC - two body system constrained to oscillate in heave

HSSV simulation requires a new definition of Ψ

System of equations written as a 1st-order ODE

x are called the states

Maximization of a performance index based on the control u of the HSSV

Using the Pontriagyn Maximum Principle (PMP) the optimal problem is

λ 0 s are called adjoint variables

The PMP shows that:

H (t,x, u, λ) = λT f (t,x,u) + L (x, u) ,

is maximum for the optimal input u subjected to:

non-causal control - Fd (t) required to compute f for t ∈ [ 0,Tf ]

See [3, 4] for further details

The algorithm sometimes opens/closes the HSSV intermittently during short

A new challenge in the HSSV control

The sub-optimal problem of open/closing the HSSV only at the end of

No solution found in the searched bibliography!

We start the quest for a new solution method...

The computational domain is discretized in small time-elements

The numerical solution is continuous within the time elements and

Solution of the sub-optimal control problem - compute u that maximizes the

In the DG Finite Element Method, the original problem

is replaced by a weak formulation

OWC motion is modelled as a rigid piston (LOWC λwaves )

p AHSSV p AOWC since AHSSV AOWC