explicit model predictive control of a fuel celldrgona/files/presentation/eucco_fuell_cell.pdf ·...
TRANSCRIPT
Explicit Model Predictive Control of a Fuel Cell
Deepak Ingole, Jan Drgona, Martin Kaluz, Martin Klauco,Monika Bakosova, Michal Kvasnica
Faculty of Chemical and Food TechnologySlovak University of Technology in Bratislava
Slovakia
September 12, 2016
Acknowledgment: The research leading to these results has received funding from the People Programme (Marie CurieActions) of the EU’s Seventh Framework Programme under REA grant agreement no 607957 (TEMPO). The Authors gratefullyacknowledge the contribution of the Slovak Research and Development Agency under the project APVV 0551-11.
Deepak Ingole (STU) September 12, 2016 1 / 22
Outline
1 Fuel Cell
2 Experimental Set-up
3 PEM Fuel Cell Control
4 Fuel Cell Modeling
5 Experimental Results
6 Conclusions
Deepak Ingole (STU) eMPC of Fuel Cell September 12, 2016 2 / 22
Fuel Cell
A device which converts the chemical energy from the fuel intoelectric energy through a chemical reaction
Emits heat and pure water
Operates like a battery
Proton Exchange Membrane (PEM)
Deepak Ingole (STU) Fuel Cell September 12, 2016 3 / 22
Fuel Cell Power System
Day Night
Solar Energy
Wind Energy
Battery Bank
Electrolyzer
Water
Hydrogen
Oxygen
Consumers
FuelCell
Deepak Ingole (STU) Fuel Cell September 12, 2016 4 / 22
Motivation
Properties of fuel cells
Renewable source of energySilent operationNo pollutionHigh EfficiencyConsumer market
Deepak Ingole (STU) Fuel Cell September 12, 2016 6 / 22
Motivation
Properties of fuel cells
Renewable source of energySilent operationNo pollutionHigh EfficiencyConsumer market
Control of electrolyzer and fuel cell
Deepak Ingole (STU) Fuel Cell September 12, 2016 6 / 22
Principle of a Fuel Cell
,
Fuel In
Anode CathodeElectrolyte
Load
Oxidant In
Water Out
H2
H+
H+
H+
e− e−
O2
H2O
Deepak Ingole (STU) Fuel Cell September 12, 2016 7 / 22
Principle of a Fuel Cell
,
Fuel In
Anode CathodeElectrolyte
Load
Oxidant In
Water Out
H2
H+
H+
H+
e− e−
O2
H2O
H2 −−→ 2H+ + 2 e
−
1
2O2 + 2H
+ + 2 e−
−−→ H2O
Deepak Ingole (STU) Fuel Cell September 12, 2016 7 / 22
Components of a Fuel Cell System
FuelFuel
ProcessorHydrogen
Fuel CellStack
Air
DC AC
Electricity
Clean Exhaust
Heat and Water
PowerConverter
Deepak Ingole (STU) Fuel Cell September 12, 2016 8 / 22
Experimental Set-up of a PEM Fuel Cell
Host PC
Electrolyzer
Input DataMonitor
Tank Load
FC Stack
Output DataMonitor
Deepak Ingole (STU) Experimental Set-up September 12, 2016 9 / 22
PEM Fuel Cell Control
Generate desired voltage from the fuel cell stack without violating theelectrolyzers input voltage limits
Challenges
Electrolyzer and fuel cell dynamicsTemperature and air flowHigh resistance at load sideSmall operating range of electrolyzers
Control
Model predictive controlDisturbance modeling
Deepak Ingole (STU) PEM Fuel Cell Control September 12, 2016 10 / 22
Implementation of a Real-time MPC
Real-time MPC
Deepak Ingole (STU) Fuel Cell Modeling September 12, 2016 11 / 22
Implementation of a Real-time MPC
Plant Model Real-time MPC
Deepak Ingole (STU) Fuel Cell Modeling September 12, 2016 11 / 22
Implementation of a Real-time MPC
Plant Model Plant/Model Mismatch Real-time MPC
Deepak Ingole (STU) Fuel Cell Modeling September 12, 2016 11 / 22
Implementation of a Real-time MPC
Plant Model Plant/Model Mismatch MPC Design Real-time MPC
Deepak Ingole (STU) Fuel Cell Modeling September 12, 2016 11 / 22
PEM Fuel Cell Modeling
VOUT
VoltageRegulator
VIN
FC Stack LoadElectrolyzerCanister
+
−
Fuel Cell Plant
Hydrogen
Input: Input voltage (VIN)
Output: Output voltage (VOUT)
Input Constraints: 1.8-2.12 V
Deepak Ingole (STU) Fuel Cell Modeling September 12, 2016 12 / 22
Model Identification
0 100 200 300 400 500 600 700 800 900
1.86
1.88
1.9
1.92
Time [s]
InputVoltage
[V]
0 100 200 300 400 500 600 700 800 900
1.2
1.4
1.6
1.8
2
Time [s]
OutputVoltage
[V]
Deepak Ingole (STU) Fuel Cell Modeling September 12, 2016 13 / 22
Model Identification
0 100 200 300 400 500 600 700 800 900
1.86
1.88
1.9
1.92
Time [s]
InputVoltage
[V]
0 100 200 300 400 500 600 700 800 900
1.2
1.4
1.6
1.8
2
Time [s]
OutputVoltage
[V]
Deepak Ingole (STU) Fuel Cell Modeling September 12, 2016 13 / 22
Model Validation
Model Order Fit To Estimated Data FPE MSE
1 95.6 % 3.9×10−5 3.9×10−5
2 96.5 % 2.4×10−5 2.3×10−5
3 96.6 % 2.3×10−5 2.2×10−5
4 96.7 % 2.2×10−5 2.1×10−5
5 96.8 % 2.1×10−5 2.0×10−5
Identified fuel cell model:
xk+1 = Axk + Buk
yk = Cxk
Deepak Ingole (STU) Fuel Cell Modeling September 12, 2016 14 / 22
Disturbance Modeling
Sources of model/plant mismatch
Temperature and air flowData monitor
Augmented design model
xk+1 = Axk + Buk
dk+1 = dk
yk = Cxk + dk
Luenberger observer
[
x
d
]
k+1
=
[
A 00 I
] [
x
d
]
k
+
[
B
0
]
uk + L (ym,k − yk)
yk =[
C I]
[
x
d
]
k
Deepak Ingole (STU) Fuel Cell Modeling September 12, 2016 15 / 22
MPC Problem
minu0,...,uN−1
N−1∑
k=0
(yk − yr )TQ(yk − yr ) + ∆uTk R∆uk
s.t. xk+1 = Axk + Buk k = 0, . . . ,N − 1
yk = Cxk k = 0, . . . ,N − 1
∆uk = uk − uk−1 k = 0, . . . ,N − 1
uk ∈ U k = 0, . . . ,N − 1
x0 = x(t)
Deepak Ingole (STU) Fuel Cell Modeling September 12, 2016 16 / 22
MPC Problem
minu0,...,uN−1
N−1∑
k=0
(yk − yr )TQ(yk − yr ) + ∆uTk R∆uk
s.t. xk+1 = Axk + Buk k = 0, . . . ,N − 1
yk = Cxk + dk k = 0, . . . ,N − 1
∆uk = uk − uk−1 k = 0, . . . ,N − 1
uk ∈ U k = 0, . . . ,N − 1
dk+1 = dk k = 0, . . . ,N − 1
x0 = x(t)
d0 = d(t)
Deepak Ingole (STU) Fuel Cell Modeling September 12, 2016 16 / 22
Explicit MPC
Optimization
Problem (mpQP)
Deepak Ingole (STU) Fuel Cell Modeling September 12, 2016 17 / 22
Explicit MPC
Optimization
Problem (mpQP)
Explicit Solution
(Look-Up Table)
Estimation
Off-line
On-line
Plant
u⋆(t)
y(t)
x(t), d(t)
Deepak Ingole (STU) Fuel Cell Modeling September 12, 2016 17 / 22
Sequential Search
u⋆(x)
x
R3 R4 R5 R6 R7R2R1
A1x ≤ b1 A2x ≤ b2 A3x ≤ b3 A4x ≤ b4 A5x ≤ b5 A6x ≤ b6 A7x ≤ b7
u⋆(x) =
F1x0 + g1 if x0 ∈ R1
...
F7x0 + g7 if x0 ∈ R7
Deepak Ingole (STU) Fuel Cell Modeling September 12, 2016 18 / 22
Sequential Search
u⋆(x)
x
R3 R4 R5 R6 R7R2R1
A1x ≤ b1 A2x ≤ b2 A3x ≤ b3 A4x ≤ b4 A5x ≤ b5 A6x ≤ b6 A7x ≤ b7
u⋆(x) =
F1x0 + g1 if x0 ∈ R1
...
F7x0 + g7 if x0 ∈ R7
Deepak Ingole (STU) Fuel Cell Modeling September 12, 2016 18 / 22
Sequential Search
u⋆(x)
x
R3 R4 R5 R6 R7R2R1
A1x ≤ b1 A2x ≤ b2 A3x ≤ b3 A4x ≤ b4 A5x ≤ b5 A6x ≤ b6 A7x ≤ b7
u⋆(x) =
F1x0 + g1 if x0 ∈ R1
...
F7x0 + g7 if x0 ∈ R7
Deepak Ingole (STU) Fuel Cell Modeling September 12, 2016 18 / 22
Sequential Search
u⋆(x)
x
R3 R4 R5 R6 R7R2R1
A1x ≤ b1 A2x ≤ b2 A3x ≤ b3 A4x ≤ b4 A5x ≤ b5 A6x ≤ b6 A7x ≤ b7
u⋆(x) =
F1x0 + g1 if x0 ∈ R1
...
F7x0 + g7 if x0 ∈ R7
Deepak Ingole (STU) Fuel Cell Modeling September 12, 2016 18 / 22
Explicit MPC : Pros and Cons
Pros
© Simple implementation: small code, division-free© Predictable execution: exact worst-case runtime and memory© Verifiable performance: closed-loop stability, feasibility, and safety
Cons
§ Only for small scale problems§ Explicit solutions can be very complex§ Reducing complexity requires scarifying performance
Deepak Ingole (STU) Fuel Cell Modeling September 12, 2016 19 / 22
Implementation of a Explicit MPC
Prediction horizon: 10
Number of regions: 519
Computational time: 3.58 s u⋆(x)
x2
x1
Deepak Ingole (STU) Experimental Results September 12, 2016 20 / 22
Experimental Results
100 200 300 400 500
1.7
1.75
1.8
1.85
OutputVoltage
[V]
Reference
100 200 300 400 5001.8
1.9
2
2.1
Time [s]
InputVoltage
[V]
Lb/Ub
Deepak Ingole (STU) Experimental Results September 12, 2016 21 / 22
Experimental Results
100 200 300 400 500
1.7
1.75
1.8
1.85
OutputVoltage
[V]
ReferenceOutput
100 200 300 400 5001.8
1.9
2
2.1
Time [s]
InputVoltage
[V]
Lb/UbInput
Deepak Ingole (STU) Experimental Results September 12, 2016 21 / 22