1. Model predictive control for energy efficient cooling and dehumidification
Tea Zakula
Leslie Norford
Peter Armstrong

2. Motivation Total U.S. energy consumption Primary energy use 1
Commercial
buildings
Residential
Transportation
Industrial
kg oil equivalent/capita
average
Source: The World Bank (2010)
Source: U.S. Energy Information Administration (2012) 1
IBO Workshop, Boulder, Colorado, June 2013

3. LLCS description
Low-Lift Cooling System (LLCS) delivers cold water to Thermally Activated Building Surfaces
(TABS). Cooling is optimized by the Model Predictive Control (MPC) algorithm.
Model predictive control
Heat pump
Cold water
Building with TABS and thermal storage
Dedicated outdoor air system (DOAS)
Ventilation and
dehumidification air 2
IBO Workshop, Boulder, Colorado, June 2013

4. Model predictive control
LLCS description
Model Predictive Control (MPC) – Cooling is optimized over 24-hours for the lowest energy consumption. Building is precooled during night when the cooling process is more efficient.
Model predictive control
Heat load
predictions
Zone temperature
Temperature limits
Cool
Non-occupied
Occupied
Non-occupied
Optimized cooling 3
IBO Workshop, Boulder, Colorado, June 2013

5. LLCS savings strategies
Cooling cycle in T-s diagram T
Thermally Activated Building Surface (TABS) - increases evaporating temperature and reduces transport power.
Thermal storage – reduces condensing temperature, peak loads and daytime loads.
Use building as thermal storage saves useful building space.
Dedicated Outdoor Air System (DOAS) – provides better ventilation and humidity control.
Model Predictive Control (MPC) – enables strategic cooling, shifting cooling toward night time.
Toutisde
Tfluid s
With conventional system
With low-lift cooling technology 4
IBO Workshop, Boulder, Colorado, June 2013

6. Previous work on LLCS
Pacific Northwest National Laboratory (2009, 2010) – Proposed LLCS and assessed its performance for 16 different climates and several building types. Showed annual electricity savings up to 70%.
Gayeski’s experimental measurements (2010) – Tested LLCS in experimental room at MIT. For a typical summer week showed 25% electricity savings for Atlanta and 19% for Phoenix climate. 5
IBO Workshop, Boulder, Colorado, June 2013

7. Software environment components
Heat load
predictions
Model predictive control
Building data
Building with TABS and thermal storage
Building model
Heat pump
Cold water
Dedicated outdoor air system (DOAS)
Ventilation and
dehumidification air 6
IBO Workshop, Boulder, Colorado, June 2013

8. Data-driven (inverse) model
Building model
Building model
Data-driven (inverse) model
- Used for optimization
- Validated using TRNSYS model
TRNSYS model
- Used after the optimization to give more accurate building response
- Validated using experimental measurements 7
IBO Workshop, Boulder, Colorado, June 2013

9. Inverse building model
Inverse model of the experimental room – proposed by Armstrong (2009)
For zone, operative and floor temperature:
𝑇= 𝑘=1 𝑛 𝒂 𝑘 𝑇 𝑘 + 𝑘=0 𝑛 𝒃 𝑘 𝑇𝑜𝑢𝑡𝑖𝑠𝑑𝑒 𝑘 + 𝑘=0 𝑛 𝒄 𝑘 𝑄𝑖𝑛𝑡𝑒𝑟𝑛𝑎𝑙 𝑘 + 𝑘=0 𝑛 𝒅 𝑘 𝑄𝑐𝑜𝑜𝑙𝑖𝑛𝑔 𝑘
For water return temperature:
𝑇𝑤,𝑟𝑒𝑡𝑢𝑟𝑛= 𝑘=1 𝑛 𝒆 𝑘 𝑇𝑤,𝑟𝑒𝑡𝑢𝑟𝑛 𝑘 + 𝑘=0 𝑛 𝒇 𝑘 𝑇𝑓𝑙𝑜𝑜𝑟 𝑘 + 𝑘=0 𝑛 𝒈 𝑘 𝑄𝑐𝑜𝑜𝑙𝑖𝑛𝑔 𝑘
Tpresent
Tpast
Coefficients a … g are found using
linear regression to TRNSYS data.
Time
k=3 k=2 k=1 k=0
Toutside,past+present
Qinternal,past+present
Qcooling,past+present 8
IBO Workshop, Boulder, Colorado, June 2013

10. Software environment components
Heat load
predictions
Model predictive control
Building data
Heat pump optimization
Building with TABS and thermal storage
Heat pump
Cold water
Dedicated outdoor air system (DOAS)
Ventilation and
dehumidification air 9
IBO Workshop, Boulder, Colorado, June 2013

11. Heat pump optimization results
Physics based heat pump model (Zakula, 2010) used to find power consumption in optimal point and optimal set of parameters required to achieve that point.
Specific power consumption in the optimal point
Parameters required to
achieve the optimal point
Evaporator airflow
Condenser airflow
Compressor frequency
Subcooling on condenser
1/COP
Qcooling/Qcooling,max
Results of static optimization are used
in software environment to calculate electricity required for cooling.
Toutside = 30 oC
Tw,return = 20 oC
Tw,return = 17 oC
Tw,return = 14 oC
Tw,return = 11 oC 10
IBO Workshop, Boulder, Colorado, June 2013

12. Software environment components
Model predictive control
Heat load
predictions
Model predictive control
Building data
Building with TABS and thermal storage
Heat pump
Cold water
Dedicated outdoor air system (DOAS)
Ventilation and
dehumidification air 11
IBO Workshop, Boulder, Colorado, June 2013

13. Model predictive control
Find the optimal cooling rates for the lowest electricity consumption over the planning horizon.
Planning horizon
Qc0
Qc1 Qc Qc23
Cooling rate optimization
Time (h)
Objective funtion= i=0 23 Cooling power+ i=0 23 Transport power + i=0 23 Temperature penalty
Using heat pump optimization results
Using inverse building model 12
IBO Workshop, Boulder, Colorado, June 2013

14. Model predictive control
Qc … cooling rate
Tw … water temperature
To … operative temperature
Tz … room temperature
Tfloor … floor temperature
Optimization
MATLAB
Optimization of cooling rates
Building response from inverse model
Cooling electricity consumption from heat pump static optimization results
Building response
Tz,history, To,history,
Tfloor,history,Tw,history
Execution
TRNSYS
Building
thermal
response
Optimal values
[Qc0, Qc2, …, Qc23] 13
IBO Workshop, Boulder, Colorado, June 2013

15. Model predictive control main findings
Software environment can be used for the LLCS analysis, but also for the analysis of other heating and cooling systems that employ MPC.
Building with
LLCS
MPC
VAV
split-system
Zone temperature (oC)
Time (h)
Inverse model can adequately replicate results from TRNSYS.
Inverse model
TRNSYS
Model is fast enough for implementation in a real building (computational time to optimize one week is 5 – 10 min). 14
IBO Workshop, Boulder, Colorado, June 2013

16. Software environment components
Heat load
predictions
Model predictive control
Building data
Building with TABS and thermal storage
Heat pump
DOAS configurations
Cold water
Dedicated outdoor air system (DOAS)
Ventilation and
dehumidification air 15
IBO Workshop, Boulder, Colorado, June 2013

17. Proposed DOAS configurations
Enthalpy wheel
System A
Evaporator
Condenser
Enthalpy wheel
System B
Evaporator
Condenser
Enthalpy wheel
System C
Evaporator
Condenser
Enthalpy wheel
System D
Evaporator
Condenser
Enthalpy wheel
System E
Condenser
Run-around
heat pipe
Evaporator 16
IBO Workshop, Boulder, Colorado, June 2013

18. 17 LLCS vs conventional VAV LLCS vs VAV with MPC
LLCS vs conventional split-system
LLCS vs split-system with MPC 17
IBO Workshop, Boulder, Colorado, June 2013

19. LLCS vs conventional VAV
Simulating a typical summer week and 22-week period across 16 climates assuming standard internal loads (from people and equipment) for an office.
LLCS
VAV
Condenser
Condenser
DOAS
Fresh air for ventilation and dehumidification
Evaporator
Evaporator
Air for cooling, ventilation and dehumidification
Water for cooling
Operated under MPC
with temperatures allowed to float
between 20 and 25oC during occupied hours
Operated under conventional control
(only during the operating hours to
maintain constant temperature of 22.5oC) 18
IBO Workshop, Boulder, Colorado, June 2013

20. DOAS configurations analyzed with LLCS
Enthalpy wheel
System A
Evaporator
Condenser
Enthalpy wheel
System B
Evaporator
Condenser
Enthalpy wheel
System C
Evaporator
Condenser
Enthalpy wheel
System D
Evaporator
Condenser
Enthalpy wheel
System E
Condenser
Run-around
heat pipe
Evaporator 19
IBO Workshop, Boulder, Colorado, June 2013

21. LLCS vs conventional VAV
Results: zone temperatures and cooling rates for Phoenix climate
LLCS under MPC
Conventional VAV
Temperature (oC)
Time (h)
Temperature (oC)
Time (h)
Temperature limits
Operative temperature
Thermal load (W)
Thermal load (W)
Time (h)
Time (h)
Internal sensible gain
Internal sensible gain
TABS cooling rate
VAV cooling rate
DOAS cooling rate 20
IBO Workshop, Boulder, Colorado, June 2013

22. LLCS vs conventional VAV
Results: LLCS electricity savings for a typical summer week
Electricity savings (%)
→ typical and best performing
A LLCS with condenser placed outside
C LLCS with parallel condensers, one in supply, the other in return stream
D LLCS with parallel condensers, one in supply stream, the other outside
E LLCS with condenser placed outside and run-round heat pipe
→ second best performing 21
IBO Workshop, Boulder, Colorado, June 2013

23. Model predictive control
VAV under MPC
Heat load
predictions
Model predictive control
Building data
Heat pump
Cold air 22
IBO Workshop, Boulder, Colorado, June 2013

24. LLCS vs VAV under MPC
Simulating a typical summer week and 22-week period across 16 climates assuming standard internal loads (from people and equipment) for an office.
LLCS
VAV
Condenser
Condenser
DOAS
Fresh air for ventilation and dehumidification
Evaporator
Evaporator
Air for cooling, ventilation and dehumidification
Water for cooling
Operated under MPC
with temperatures allowed to float
between 20 and 25oC during occupied hours
Operated under MPC
with temperatures allowed to float
between 20 and 25oC during occupied hours 23
IBO Workshop, Boulder, Colorado, June 2013

25. LLCS vs VAV under MPC
Results: zone temperatures and cooling rates for Phoenix climate
LLCS under MPC
VAV under MPC
Temperature (oC)
Time (h)
Temperature (oC)
Time (h)
Temperature limits
Operative temperature
Thermal load (W)
Thermal load (W)
Time (h)
Time (h)
Internal sensible gain
Internal sensible gain
TABS cooling rate
VAV cooling rate
DOAS cooling rate 24
IBO Workshop, Boulder, Colorado, June 2013

26. LLCS vs VAV under MPC
Results: LLCS electricity savings for a typical summer week*
Electricity savings (%)
Results: LLCS electricity savings from May 1st – September 30th*
Electricity savings (%)
*LLCS assumes simple DOAS (system A) 25
IBO Workshop, Boulder, Colorado, June 2013

27. Lower electricity consumption Operated under conventional control
LLCS vs conventional split-system
Simulating a typical summer week in Atlanta and Phoenix, and taking into account only sensible cooling (no ventilation and dehumidification system).
LLCS
Split-system
Condenser
Condenser
Evaporator
Lower electricity consumption
33% for Atlanta
36% for Phoenix
Evaporator
Water for cooling
Operated under MPC
with temperatures allowed to float
between 20 and 25oC during occupied hours
Operated under conventional control
(only during the operating hours to
maintain constant temperature of 22.5oC) 26
IBO Workshop, Boulder, Colorado, June 2013

28. Model predictive control
Split-system under MPC
Model predictive control
Heat load
predictions
Heat pump
Heat pump 27
IBO Workshop, Boulder, Colorado, June 2013

29. Lower electricity consumption
LLCS vs split-system under MPC
Simulating a typical summer week in Atlanta and Phoenix, and taking into account only sensible cooling (no ventilation and dehumidification system).
LLCS
Split-system
Condenser
Condenser
Evaporator
Lower electricity consumption
19% for Atlanta
11% for Phoenix
Evaporator
Water for cooling
Operated under MPC
with temperatures allowed to float
between 20 and 25oC during occupied hours
Operated under MPC
with temperatures allowed to float
between 20 and 25oC during occupied hours 28
IBO Workshop, Boulder, Colorado, June 2013

30. Summary of main findings
LLCS saved up to 50% electricity relative to the VAV system under conventional control and up 23% electricity relative to the VAV system under MPC.
A split-system under MPC can have lower electricity consumption than LLCS.
Precooling had important effect for the LLCS. When allowed to precool, LLCS saved up to 20% electricity than otherwise.
Precooling did not have notable effect on the VAV system electricity consumption.
Internal loads, pipe spacing, and heat pump sizing have a significant impact on LLCS savings potential.
A typical DOAS configuration used least amount of electricity.
IBO Workshop, Boulder, Colorado, June 2013

31. Thank you!
IBO Workshop, Boulder, Colorado, June 2013