Vous êtes sur la page 1sur 28

Model Predictive Control for Central Plant Optimization

with Thermal Energy Storage

Michael J. Wenzel, Robert D. Turney, Kirk H. Drees


Johnson Controls Inc.
Outline

 Background
 Problem Statement
 Solution Framework
 Simulation Results

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 2
U.S. Energy Consumption

 Energy use for heating and


cooling in buildings
represents approximately
$250 billion in expenditures

 Large university campuses


can spend more than $5
million annually on Heating
and Cooling.

2011 Buildings Energy Data Book, March 2012, D&R International, U.S. DOE

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 3
Modern Plant Design

 Many different types of central energy plants and numerous device to choose from.
» Chillers
» Boilers
» Hot Water Generators
» Steam Turbine
» Electrical Storage
» Absorption Chillers
» Heat Pump Chillers
» Thermal Energy Storage
» Ground Source Heat Exchangers

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 4
Modern Plant Design

 What is the best plant configuration for your campus?

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 5
Modern Plant Design

 What is the best plant configuration for your campus?


» With all the different possible plant configurations and devices there are too many variables
for a human to decide.
» Regional rules of thumb.

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 6
Modern Plant Design

 What is the best plant configuration for your campus?


» With all the different possible plant configurations and devices there are too many variables
for a human to decide.
» Regional rules of thumb.

 What is the best operating point (trajectory) for my central plant?


» Many devices and setpoints to choose from.
» Possible use of thermal energy storage means that the devices and setpoints must be chosen
for the entire forward looking horizon.

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 7
Modern Plant Design

 What is the best plant configuration for your campus?


» With all the different possible plant configurations and devices there are too many variables
for a human to decide.
» Regional rules of thumb.

 What is the best operating point (trajectory) for my central plant?


» Many devices and setpoints to choose from.
» Possible use of thermal energy storage means that the devices and setpoints must be chosen
for the entire forward looking horizon.

Do not know the answer, but we could design a computer algorithm.

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 8
Plant Optimization Problem

How does a computer understand this?

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 9
Plant Optimization Problem

It is necessary to describe the problem in a way a computer can solve it.


 Cascaded Optimization
» Subplant Level Optimization – Divides the load between the various types of equipment and
the thermal energy storage tanks.
» Equipment Level Optimization – Determines which of the equipment within a subplant to run
and at what setpoints.
 Mathematical Framework
» Linear Programming – To quickly solve the subplant dispatch problem with a large horizon.
» MILP (NLP) – To solve the equipment level problem when there is no horizon.

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 10
Plant Optimization Problem

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 11
Plant Optimization Problem

 Equipment Level Optimization Rate of utility use j


by device i
Uncontrolled conditions

 nu 
       
ne

 *
Q ,  w  arg min J LL  LL , Q ,  w J LL  LL , Q, w    c j u ji  LL , Q , w 

 j 1 
LL
 LL i 1 

Decisions (setpoints, on/off)


Subplant Load Cost of utility j

 Subplant Level Optimization


Rate of utility use j by
subplant i at time k

 nu 
 arg min J HL  HL 
nh ns
 HL J HL  HL     c jk u jik  HL 
*

k 1 i 1  
 HL
 j 1

Decisions (load distribution)


Cost of utility j at
time k

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 12
Proposed Solution

 Linear Programming Framework


» Good method to quickly handle the large number of variables of the subplant level
optimization problem.
– 168 hour horizon × 5 decisions = 840 decision variables in the horizon (for the simplest problem).

» Seems restrictive at first, but Possible to program:


– Demand Charges
– Load Change Penalties
– Plant Performance Curves (Convex)

Decisions (load distribution)

arg min c T x; subject to Ax  b, Hx  g ,


x

Cost vector

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 13
Proposed Solution

Central Plant Optimization as a Linear Program


 Decision Variables
» The load on each subplant and storage tank.


x  Q chiller,1...n , Q hrChiller,1...n , Q heater,1...n , Q hotStorage,1...n , Q coldStorage,1...n 
T

 Cost Vector
» Simple case (assuming constant cost per load)

T
 n  n  n  
c    c j u j ,chiller  ,  c j u j ,hrChiller  ,  c j u j ,heater  ,0 h ,0 h 
u u u

  j 1  1...h  j 1  1...h  j 1  1...h 


 

Must sum all utility use to get total cost

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 14
Proposed Solution

Capacity Constraints
 Subplant Capacity
» One set for each subplant
Q chiller, k  Q chiller, max k  horizon,
Q chiller, k  0 k  horizon.

» Matrix Form
 I  0 h  0 h  0 h  0 h  Q chiller, max 1h 
A h   ,
0 h  0 h  0 h  0 h 
, b
 I h   0 h  

 Tank (Dis)charge Rates


» One for each tank

0  0 h  0 h  I h  0 h  Q discharge, max 1h 


A h    ,
0 h  0 h   I h  0 h 
, b
0 h   Qcharge, max 1h  

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 15
Proposed Solution

Capacity Constraints Background campus electrical demand


 Electrical Demand Constraint


A  u electrical,chillerI h , u electrical,hrChiller I h , u electrical, heater I h ,0 n ,0 n  , bP
elec, max 1h  Pelec,campus,k

 Tank Capacities
» One for each tank

0  0 h  0 h  Ts  h  0 h   Q0, Hot 1h  


A h  , b ,
0 h  0 h  0 h  Ts   h  0 h  Q max,Hot  Q 0 , Hot 
1 h 

Lower Triangular Matrix

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 16
Proposed Solution

Load Constraints
 Must meet loads with equality.

I h  I h  0 h  0 h  I h  lˆ 
H  g   Cold ,1..k .
0 h  1  u electrical, hrChiller I h  I h  I h  0 h 
,
 lˆHot ,1...k 

 Constraints: Summary
Load Constraints (2 Loads) * (h Horizon) = 2h Constraints
Capacity Constraints (2 per Device) * (5 Devices) * (h Horizon) = 10h Constraints
Demand Constraint (h Horizon) = 1h Constraints
Tank Constraints (2 per Tank) * (2 Tanks) * (h Horizon) = 4h Constraints

17h Constraints

168 hour horizon × 17 constraints = 2856 constraints (for the simplest problem).

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 17
Proposed Solution

Demand Charge
 Non Linear Problem


arg min c T x  c demand max Pelec, k ( x) ,
x

 Make Linear using Constraints


» New peak demand decision variable


c new  c T , c demand
T
,
T

x new  x T , x peak 

» Constraint to force xpeak = max(Pelec,k(x)).


A  uelectrical,chillerI h , uelectrical, hrChiller I h , uelectrical, heater I h ,0 n ,0 n ,1h  , b  P
elec, campus , k

 Weight the cost of demand because the energy and demand charge cover different
periods.

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 18
Simulation Results

Simulation
 Given:
» Year of utility rates
» Year of central plant loads
» Year of background electric
loads
 Run in “Planning Mode”

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 19
Simulation Results

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 20
Simulation Results

Charge tanks during the


night when prices are
cheap.

Discharge during the Day

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 21
Simulation Results
Chillers produce 100% of
load for an hour, then the
tanks are discharged, then
the chillers are on again.

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 22
Simulation Results

Change of load penalty


added.

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 23
Simulation Results

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 24
Simulation Results

Demand is at least 47MW


for all months and 54MW in
the Summer.

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 25
Simulation Results

With demand charge in the


optimization algorithm
peak demand is 43 MW.

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 26
Simulation Results

hrzn blck elec. elec. elec. elec. natural natural water water Total Com
energy energy demand demand gas energy gas cost (km3) cost Cost p.
(GWh) cost (M$) (MW) cost (M$) (GWh) (M$) (M$) (M$) Time
(s)

24 24 96.3 5.51 5.35 2.69 72.2 1.33 98.5 0.25 9.78 32.9
24 12 96.4 5.50 5.27 2.69 71.7 1.32 98.1 0.25 9.76 62.3
48 12 97.0 5.54 52.6 2.68 69.0 1.27 96.3 0.24 9.75 162
72 12 97.2 5.55 52.6 2.70 67.9 1.25 95.6 0.24 9.74 431
96 8 97.3 5.56 52.6 2.70 67.4 1.24 95.3 0.24 9.74 1199
168 1 97.4 5.56 52.6 2.70 67.2 1.24 95.1 0.24 9.74 25937

INCLUDE DEMAND CHARGE OPTIMIZATION------------------------------------------------------------------------------


96 8 97.1 5.60 42.5 2.23 68.7 1.27 96.1 0.24 9.34 1236
INCLUDE LOAD CHANGE PENALTY $1 PER PERCENT PER HOUR--------------------------------------------------
96 8 97.4 5.60 48.9 2.60 67.3 1.24 95.1 0.24 9.68 1430
NO OPTIMIZATION-----------------------------------------------------------------------------------------------------------------
N/A N/A 75.3 4.44 56.1 2.63 171 3.15 170 0.43 10.65 N/A

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 27
Summary Conclusion

 Cascaded approach makes optimal use of


computational time.
» No horizon at the equipment level (fast dynamics)
» Long horizon at the subplant level (manage large
storage)
 Linear programming framework is a efficient way of
solving the subplant level problem.
 Savings of 10 – 15% over a schedule based system.
 Savings of ~4% attributed to the demand charge.

July 16, 2014 Model Predictive Control for Central Plant Optimization with Thermal Energy Storage 28

Vous aimerez peut-être aussi