an integrated framework for parametric design using building … · 2020. 7. 2. · michael...

An integrated framework for parametric design

using building energy models

Bryan Eisenhower

Associate Director

Center for Energy Efficient Design

UCSB

LBL

September 22, 2011

Motivation: Comfort & energy models are refined but require a lot

of partially certain user input

Uncertainty / sensitivity analysis supports

optimization, calibration, and FMEA.

The study of building dynamics is needed for efficient

and effective controller design

System level dynamic modeling for control analysis

and Hardware in the Loop studies

Buildings produce enormous amounts of data, too

much to analyze

Spectral / modal decomposition of BMS data

Motivation / Topics: Comfort & energy models are refined but require a lot

of partially certain user input

Uncertainty / sensitivity analysis supporting

optimization, model calibration, and FMEA.

The study of building dynamics is needed for efficient

and effective controller design

System level dynamic modeling for control analysis

and Hardware in the Loop studies

Buildings produce enormous amounts of data, too

much to analyze

Spectral / modal decomposition of BMS data

0.5 1 1.5

x 106

0

100

200

300

400

Fre

qu

en

cy

Total Power

Seasonal Consumption - Cooling

0.5 1 1.5 2

x 106

0

100

200

300

400

Fre

qu

en

cy

Total Power

Seasonal Consumption - Heating

-1.5 -1 -0.5 00

100

200

300

Fre

qu

en

cy

PMV Avg.


-2 -1.5 -1 -0.5 00

100

200

300

Fre

qu

en

cy

PMV Avg.

PMV Avg. - Heating

Sensitivity

Decomposition

methods

Energy Visualization

Uncertainty Analysis

Advanced Energy Modeling

Data analysis toolkits

Energy/Comfort

Optimization Failure Mode Effect Analysis

Summary

1 1.1 1.2 1.3 1.4 1.5 1.6 1.7

x 1011

0

50

100

150

200

250

Pumps [J] - Yearly Sum

Fre

quency

0.9 0.95 1 1.05 1.1 1.15 1.2 1.25

x 108

0

20

40

60

80

100

120

140

160

180

Interior Lighting [J] - Yearly Peak

Fre

quency

0.5 1 1.5 2 2.5 3 3.5

x 1011

0

100

200

300

400

500

600

700

Heating [J] - Yearly Sum

Fre

quency

1600 1650 1700 1750 1800 1850 1900 1950 20000

20

40

60

80

100

120

Occurr

ences

VAV3 Availability Manager

0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.090

50

100

150

200

250

300

Occurr

ences

BLDGLIGHT

SCH

Uncertain Inputs

1600 1650 1700 1750 1800 1850 1900 1950 20000

20

40

60

80

100

120

Occurr

ences


Building Model

Uncertain Outputs

?

1.1 1.2 1.3 1.4 1.5 1.6

x 104

1.6

1.7

1.8

1.9

2

2.1

2.2

2.3

2.4x 10

4 Month 3

Plug Electricity [kW]

Tota

l E

lectr

icity [

kW

]

Sampled Output

Sensor Data

Baseline Model

Calibrated Meta

Calibrated E+

Model Calibration

0 0.1 0.2 0.3 0.4

Lighting not turned off at night

Nightsetpoint temperature set incorrectly

Zone 7 Thermostat improperly located

Boiler gas/air flow restricted/leaks

AHU2 Economizer OA damper fails open

Ballast Efficiency Degradation

Zone 6 lights left on

AHU1 Min OA damper fails open

Sensitivity Index

Output Num. 15 total.sum

Acknowledgements

This work was partially supported under

the contract W912HQ-09-C-0054 Project Number: SI-1709

administered by SERDP technology program of the Department

of Defense.

Collaborators:

Zheng O’Neill United Technologies Research Center

Satish Narayanan United Technologies Research Center

Shui Yuan United Technologies Research Center

Vladimir Fonoberov AIMdyn Inc.

Kevin Otto RSS

Igor Mezic University of California, Santa Barbara

Michael Georgescu, Erika Eskenazi, Valerie Eacret UCSB

Analysis leveraging UA / SA

Identify uncertain parameters, perform

sampling

Perform numerous simulations, pre-process

output

Calculate full order meta-model

Calculate reduced-order

meta-model

Perform UA/SA

Confirm optimal with full

simulation

Perform Optimization


Create Energy Model E+, TRNSYS, Modelica

Identify Mechanisms for Failure Modes, influences on parameters


output


Identify most significant Failure Modes


simulation

Define decomposition

structure

Perform UA/SA for

dec. system


meta-model

Perform UA/SA

Perform Calibration/Optim

ization

Verify Calibration Perform UA/SA

Uncertainty / Sensitivity Analysis Optimization Calibration

Perform UA/SA

FMEA

Flow


sampling


output



meta-model

Perform UA/SA


simulation




Identify Mechanisms for Failure Modes, influences on parameters


output


Identify most significant Failure Modes


simulation


structure

Perform UA/SA for

dec. system


meta-model

Perform UA/SA


ization

Verify Calibration Perform UA/SA

Uncertainty / Sensitivity Analysis Optimization Calibration

Perform UA/SA

FMEA

3-6 weeks

2 human days, 1.5 calendar

weeks

1-2 weeks

2 human days, 1.5 calendar

weeks

Scalability

Case studies

Case Studies

Building 1225 in Ft. Carson with TRNSYS

An administration and training facility built in 70’s.

One floor with an area of ~24000 ft2.

Major HVAC systems: 2 constant-air-volume

multi-zone-units, chilled water from a central

plant May-October, hot water by a gas boiler

November-April.

Domestic hot water generated by a gas water

heater.

DOE benchmark models

Medium office model in Las Vegas

3 floors, ~50K ft^2, 15 zones

5min/simyear on Linux cluster

30min/simyear on windows server

DOD: Atlantic Fleet Drill Hall

6430 m2 69 K ft^2

Model developed in EnergyPlus

30 Conditioned zones

1009 uncertain parameters

Case Studies

25min/simyear on Linux

Building 26, Fleet and Family Support Center FFSC/Navy Marine Corps Relief Society NMCRS

•Naval Training Center, Great Lakes, IL

•Two-storey office building with basement.

•The gross area of this building is approximately 37,000 ft2

•Two airside systems and two separated waterside systems

•The office and administrative area on the first and second floors is served by two variable-air volume

VAV Air Handling Units, AHU with VAV terminal unit with hot water reheat heating and cooling

capability.

•These AHUs have both heating and cooling capability

•The chilled water system consists of one 54.5-ton air-cooled rotary-screw type chillers with fixed-

speed primary pumping.

•Heating is supplied from the existing base-wide steam system through a steam-to-water heat

exchanger

•The hot water serves unit heaters, VAV box reheating coils, and air handling unit heating coils.

•The communication service room is served by one dedicated split system.

•Electric unit heater and baseboard are used to provide heating to stairwells and restrooms.

• Modeled in: EnergyPlus v6.0 • 1 year sim ~30 Min on 2.8GHz processor

30min/simyear on Linux

Case Studies

Discrepancy is often introduced because of uncertainty

Commissioning / Operation

Material selection

Usage

… Other unknowns

Energy Modeling & Uncertainty

Sensitivity / Uncertainty Analysis

Discrepancy is often introduced because of uncertainty

Commissioning / Operation

Material selection

Usage

… Other unknowns

Sensitivity / Uncertainty Analysis helps manage these

concerns


1 1.1 1.2 1.3 1.4 1.5 1.6 1.7

x 1011

0

50

100

150

200

250

Pumps [J] - Yearly Sum

Fre

quency

0.9 0.95 1 1.05 1.1 1.15 1.2 1.25

x 108

0

20

40

60

80

100

120

140

160

180

Interior Lighting [J] - Yearly Peak

Fre

quency

0.5 1 1.5 2 2.5 3 3.5

x 1011

0

100

200

300

400

500

600

700


Fre

quency

1600 1650 1700 1750 1800 1850 1900 1950 20000

20

40

60

80

100

120

Occurr

ences


0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.090

50

100

150

200

250

300

Occurr

ences

BLDGLIGHT

SCH

Uncertain Inputs

1600 1650 1700 1750 1800 1850 1900 1950 20000

20

40

60

80

100

120

Occurr

ences


Building Model

Uncertain Outputs

?

O1000 O10

Sampling

• O.A.T.

• Monte Carlo

• Latin Hypercube

• Quasi-Monte Carlo

deterministic

• All parameters at once


Red: In this talk

Uncertainty Analysis

• STD, VAR

• COV

• Amplification

factors

Sensitivity Analysis

• Elementary Effects /

screening & local methods

• Morris Method

• ANOVA

• Derivative-based

• Propagation analysis

through decomposition

Disclaimer: The methods here are what is historically done in building systems, and what we have applied in our work. Other methods exist that are not discussed here poly-chaos, waveform relaxation, etc.

UA / SA – Historically Building Sys.

Authors # Param. Technique Notes

Rahni [1997] 390->23 Pre-screening

Brohus [2009] 57->10 Pre-screening / ANOVA

Spitler [1989] 5 OAT / local Residential housing

Struck [2009] 10

Lomas [1992] 72 Local methods

Lam [2008] 10 OAT 10 different building types

Firth [2010] 27 Local Household models

de Wit [2009] 89 Morris Room air distribution model

Corrado [2009] 129->10 LHS / Morris

Heiselberg [2009] 21 Morris Elementary effects of a building model

Mara [2008] 35 ANOVA Identify important parameters for calibration also.

Capozzoli [2009] 6 Architectural parameters

Eisenhower [2011] 1009 , to 2000

Deterministic sampling, global derivative sensitivity

‘All’ available parameters in building

UA / SA Results

Uncertainty analysis investigates the

forward influences of parameter

variance

UA / SA Results

Uncertainty analysis investigates the

forward influences of parameter

variance

Sensitivity analysis identifies which of

the parameters are most influential


sampling


output




structure

Perform UA/SA for

dec. system

Perform UA/SA

Uncertainty / Sensitivity Analysis

Flow

Parameter Variation

All numerical design & operation scenario (DOS) parameters in the

model are varied concurrently (not architectural design)

Parameter Type Examples

Heating source Furnace, boiler, HWGSHP etc

Cooling source chiller, CHWGSHP etc

AHU AHU SAT setpoint, coil parameters etc

Air Loop Fans

Water Loop Pumps

Terminal unit VAV box, chilled beam, radiant heating floor

Zone external Envelope, outdoor conditions

Zone internal Usage, internal heat gains schedule,

Zone setpoint Zone temp setpoint

Sizing parameter Design parameters for zone, system, plant

1600 1650 1700 1750 1800 1850 1900 1950 20000

20

40

60

80

100

120

Occurr

ences


0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.090

50

100

150

200

250

300

Occurr

ences

BLDGLIGHT

SCH

nominal 20-25%

Parameters varied 20-25% of their mean

Some parameters are of the form a+b < 1

Distribution types are

available in literature

but not applied

because of the large

number of parameters

Parameter Variation

Large number of parameters and lengthy simulation time require

efficient parameter selection for parameter sweeps

Deterministic sampling provides uniformly ergodic samples

Random

Deterministic

Typical Output Distributions

Facility and Submetered Outputs

+ Gas Facility

+ Electricity Facility

Heating

Cooling

Pump

Fan

Interior Lighting

Interior Equipment

5.3 5.4 5.5 5.6 5.7 5.8 5.9 6

x 1011

0

50

100

150

200

250

300

350

Interior Lighting [J] - Yearly Sum

Fre

quency

* TRNSYS results

0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

x 1012

0

50

100

150

200

250

300

350


Fre

quency

5000-6000 realizations performed to obtain convergence

The ‘control’ mechanisms in the model drive distributions towards Gaussian although others exist as well

Convergence Properties

Monte Carlo bound ~ 1/sqrtN

Deterministic bound ~ 1/N

Example Convergence from Building Simulation

Faster

convergence

means more

parameters can be

studied in the

same amount of

time!

Both

deterministic,

slope ~ -0.7

Model Results - UA

Influence of Different Parameter

Variation size

[E+ Drill Hall]

Characteristics of the output based on different inputs

Input Uncertainty @ 20% Input Uncertainty @ 10%

Model Results - UA

Nominal vs. High Efficiency Design

B. Eisenhower, et al. A COMPARATIVE STUDY ON UNCERTAINTY

PROPAGATION IN HIGH PERFORMANCE BUILDING DESIGN Accepted:

International Building Performance Simulation Association, BuildSim 2011

[E+ DOE Models]

Characteristics of the output are considered based different models


sampling


output




structure

Perform UA/SA for

dec. system

Perform UA/SA


Flow

Meta-modeling - Results

A model of a model (meta-model) is created to provide

means for analytic assessment of building energy

predictions

Structure of the model is similar to the full order energy

model, not a line fit to the data.

2063 inputs, 2 outputs

http://www.da-sol.com/en/resources/starter-pages/svm-starter/

http://www.imtech.res.in/raghava/rbpred/svm.jpg Smola

[2004]

Machine Learning / Regression

Identify characteristics within data without prior

knowledge of the regressors

Applications: object detection, classification of

biological data, speech or image recognition, internet or

database searching…..

Soft margin set up to identify outliers….

Machine learner

http://www.da-sol.com/en/resources/starter-pages/svm-starter/http://www.da-sol.com/en/resources/starter-pages/svm-starter/http://www.da-sol.com/en/resources/starter-pages/svm-starter/http://www.da-sol.com/en/resources/starter-pages/svm-starter/http://www.da-sol.com/en/resources/starter-pages/svm-starter/http://www.da-sol.com/en/resources/starter-pages/svm-starter/http://www.da-sol.com/en/resources/starter-pages/svm-starter/

Lagrange multipliers

Kernel, e.g.:

Solved by

KKT

conditions

Support vector solution is obtained by solving a convex optimization

problem

Machine Learning / Regression

Derivation from [Smola 2004] * See also: www.support-vector-machines.org


sampling


output



meta-model

Perform UA/SA

Real-time implementation /

MPC


Use of SVR-based models for prediction e.g. MPC is not

new. Integration with UA would be useful.

Future possibilities….


sampling


output




structure

Perform UA/SA for

dec. system

Perform UA/SA


Flow

Sensitivity Calculation

ANOVA-based approach:

Functional decomposition

Sensitivity indices

Total individual sensitivity

Variance decomposition

12 k

terms

Sensitivity Calculation

Variance is not always best to

describe distribution

ANOVA-based approach:

Derivative-based approach:

Functional decomposition Variance decomposition

Sensitivity indices

Total individual sensitivity

Zheng O’Neill, Bryan Eisenhower, et al

Modeling and Calibration of Energy Models

for a DoD Building ASHRAE Annual Conference, Montreal 2011

Sensitivity Analysis

Identifying key parameters in a building helps in design optimization, continuous commissioning, model calibration, …

[E+ Drill Hall]

Typically only a

few parameters

drive

uncertainty in

output Hand calibration from SA

System Decomposition

http://www.biomedcentral.com/14712105/7/386/figure/F2?highres=y

What are the essential components of a productive network?

Decomposition provides an understanding of essential production units

and the pathway energy/information/uncertainty flows through the

dynamical system

Integrated Gasification Combined Cycle, or IGCC, is a technology that turns coal into gas into

electricity

Decomposition Methods

Facility

Electricity

Intermediate Consumption

Variables

Input

Parameter

Types

Uncertainty at each node and pathway flow identified for a

heterogeneous building

0 100 200 300 400 500 600 700 800 900 10000

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Parameter Index

Se

nsitiv

ity In

de

x

Electricity:Facility [J] (Annual Total)

All Data

Supply air temp setpoint

Chiller1 reference COP

Drill deck lighting schedule

AHU2 return fan maximum flow rate

AHU2 supply fan efficiency

AHU2 supply fan pressure rise

AHU1supply fan efficiency

AHU1 supply fan pressure rise

Chiller1 optimum part load ratio

Eisenhower et al. Uncertainty and Sensitivity Decomposition of Building

Energy Models Journal of Building Performance Simulation, 2011

Circles: Uncertainty at each node Line Thickness: ‘conductance’

Decomposition Methods – Building Energy

Optimization

Multi-objective Optimization

Objectives in buildings are naturally

competitive

B. Eisenhower, et al Metamodel-based

Optimization of Building Energy

Systems Submitted Warm Cold

Multi-objective Optimization

Objectives in buildings are naturally

competitive

Many whole-building energy

simulators do not lend themselves well

to optimization

Wetter & Polak 2004

Warm Cold

Optimization – Historically Building Sys.

Authors # Param. Technique Type of model / cost function

Kampf [2011] 13 Genetic, Particle swarm Full E+ model, energy only

Djuric [2007] 3 GENOPT Full E+ model, energy consumption, investment costs, comfort

Wang [2005] 27 Genetic algorithm ASHRAE sim. Program, LCC/LCEI optimized

Caldas [2003] 16 Genetic algorithm RC network model, energy and first cost

Peippo [1999] 30 Hooke Jeeves coordinated Algebraic, energy and capital

Wright [2002] 9 Genetic algorithm RC network model, energy and comfort

Gustafsson [2000] 183 MILP Algebraic, life cycle costs

Marks [1997] 20 NLP Algebraic, construction/ operation costs

Djuric [2007] 10 GENOPT Energy Plus, energy, comfort, construction

Lozano [2009] 10 LINGO Algebraic, annual energy & fixed cost

Wetter [2004] 13 GENOPT Energy Plus, annual energy

Diakaki [2008] 9 MILP, LINGO Algebraic, energy & construction

Eisenhower, submitted [2011]

1009 up to 2063

Meta-model based optimization, IPOPT & der free

E+/meta-model ‘All’ available parameters in building


sampling


output



meta-model

Perform UA/SA


simulation





simulation

Optimization

Flow


sampling


output



meta-model

Perform UA/SA


simulation





simulation


structure

Perform UA/SA for

dec. system

Perform UA/SA

Uncertainty / Sensitivity Analysis Optimization

Flow

Previous results can

certainly be used!

Optimization Strategy

Cost is typically some combination possibly weighted of comfort and energy

Two methods used:

1. IPOPT - Primal-Dual Interior Point

algorithm with a filter line-search

method for nonlinear programming

Wachter - Carnegie Melon / IBM

2. NOMAD - Derivative free Mesh

Adaptive Direct algorithm Digabel -

Ecole Polytechnique de Montreal

Works well

with analytic /

continuous

cost

Compare

with previous

work

Optimization results compared to uncertainty distributions

cooling & heating seasons

Red dot =

nominal

simulation

Green =

Maximized

solution

Blue =

Minimized

solution

Dot = 317 para

Triangle = 16

para

0.5 1 1.5

x 106

0

100

200

300

400

Fre

qu

en

cy

Total Power


0.5 1 1.5 2

x 106

0

100

200

300

400

Fre

qu

en

cy

Total Power

Seasonal Consumption - Heating

-1.5 -1 -0.5 00

100

200

300

Fre

qu

en

cy

PMV Avg.

Comfort - Cooling

-2 -1.5 -1 -0.5 00

100

200

300

Fre

qu

en

cy

PMV Avg.

Comfort - Heating

Optimization Results

Fort Carson

TRNSYS

45% annual energy reduction while increasing comfort

by a factor of two

Optimization Results

Building26

E+

Model reduction based on parameter type or parameter influence

Rank ordering of parameter sensitivity Parameters collected by type

Traditional approach

Reduced meta-

models

Full meta-model

B. Eisenhower, et al Metamodel-based

Optimization of Building Energy

Systems Submitted

Model

Calibration


sampling


output




meta-model

Perform UA/SA


ization

Verify Calibration

Calibration

Flow

Data

collection!

Calibration approach tested on DOD Building 26

Data available 2009, 2010, some 2011:

Plug electricity

Total electricity

Steam consumption

Calibration Results

2data) - model(Optimization performed to minimize Per month, year, etc.

Used for

calibration

Calibration approach tested on DOD Building 26

Data available 2009, 2010, some 2011:

Plug electricity

Total electricity

Steam consumption

1 2 3 4 5 6 7 8 9 10 11 12-10

-5

0

5

10

15

20

25

30

35

Perc

ent

Err

or

Month


Nominal

Calibrated

1 2 3 4 5 6 7 8 9 10 11 12-20

-15

-10

-5

0

5

10

15

Perc

ent

Err

or

Month

Total Electricity [kW]

Nominal

Calibrated

Calibration Results

2data) - model(Optimization performed to minimize Per month, year, etc.

Used for

calibration

1 2 3-15

-10

-5

0

5

10

15

20

25E

rror

in %

Verification with 2011 weather

Nominal with 2011 weather

Calibration with 2010 weather

March MayApril

Calibration performed on 2010 data

Verification performed with 2011 data

Results from previous slide Uncalibrated model

with new 2011 data

Calibrated model

with new 2011

data

Chiller not broken in 2010

Chiller broken in 2011

Verification Results

1.6 1.7 1.8 1.9 2 2.1 2.2

x 104

0

2

4

6

8

10

12

14


Fre

quency

January

Sampled Data

Nominal Model

Sensor Data

Calibration approach

Sensor data Un-calibrated

model

As is often the case, sensor data fall outside of

the conservative output distribution

UA distribution from

model

1.1 1.2 1.3 1.4 1.5 1.6

x 104

1.6

1.7

1.8

1.9

2

2.1

2.2

2.3

2.4x 10

4 Month 3


Tota

l E

lectr

icity [

kW

]

Sampled Output

Sensor Data

Baseline Model

Calibrated Meta

Calibrated E+

The optimization is performed while having an elliptical constraint on its

output

Constrained optimization

This is really a

circle, its

intersection with

the ellipse is as

close as we will

let the

optimization go

The meta-model and

actual E+ simulation

with optimal values

has some difference,

the meta-model is

least accurate at its

edges

To circumvent these issues, outputs are constrained, only

top 20 or so parameters varied for optimization

1 2 3 4 5 6 7 8 9 10 11 12-10

-5

0

5

10

15

20

25

30

35

Perc

ent

Err

or

Month


Nominal

Calibrated

1 2 3 4 5 6 7 8 9 10 11 12-20

-15

-10

-5

0

5

10

15

Perc

ent

Err

or

Month


Nominal

Calibrated

Calibration Results

Constrained to not be

any better than this

Do quick UA/SA to identify key

parameters and where distribution is

Use SA to hand tune the model closer to

sensor data

Perform detailed UA/SA and calibration

Future approach

Part II

Analysis of dynamics

Use of Modelica for:

1. Controller certification (testing) in a

Hardware in the Loop environment

2. Control education

Goal: Modify Carrier controller for supervisory needs

Capstone MicroTurbines

Gas -> Electricity and hot exhaust

Capstone control system

Carrier Chiller

Gas fired burner -> Cold / hot water

Carrier control system

UTRC PureComfort CHP

+

UTC PureComfort™ CHP System

Carrier Controller

Capstone Controller

…Two different control systems, one common function

Adjust to operation of chiller to different heat

source:

Micro-turbines at full power all 4 micro-

turbines on

Micro-turbines load following

All 4 micro-turbines running with power

fluctuations below 60kW

1-2 micro-turbines turned on/off with

power fluctuations greater than 60kW

All 4 micro-turbines shut down

Include Damper Valve Model

Start/Stop Procedures

Chiller does not start if all 4 micro-

turbines are turned off

Chiller shuts down safely if all 4

micro-turbines are shut down

Refine Protective Limits and Alarms

Necessary Controller Changes

…necessary changes take many

months to implement, many more to

test/certify

Adjust to operation of chiller to different heat

source:

Micro-turbines at full power all 4 micro-

turbines on

Micro-turbines load following

All 4 micro-turbines running with power

fluctuations below 60kW

1-2 micro-turbines turned on/off with

power fluctuations greater than 60kW

All 4 micro-turbines shut down

Include Damper Valve Model

Start/Stop Procedures

Chiller does not start if all 4 micro-

turbines are turned off

Chiller shuts down safely if all 4

micro-turbines are shut down

Refine Protective Limits and Alarms

Necessary Controller Changes

PureComfort

System Model

Realtime

computation

Carrier

ICVC

Controller Actuators

Sensors

HIL Experimentation Environment

~= =

…necessary changes take many

months to implement, many more to

test/certify

http://www.dspace.de/ww/en/pub/home.htm

Volumepdropv

pdropf

init charv charfgeo

charfg

burner

Q

h...

geofg initfg

f luegaspipes

a...

so...

sin...

si...

f lu...f lu...p...

c... c...

c...

co...h... chiller

p...

Ta...

of...

Di...

Di..

.

Fl...Fl...

Va...

InitV

Di...

k={1.8...

Gain1

Ma...

Mi...Mi...p...

Flow _...

PIT={10}

-

Flow ... Flow ...

uMa...

Flow ...

s...ch...

HPG

h...

LPG

HX

LX

CON

r...

hig...

c...

co...

co...

EVAABS

pump

Ejc

Bool...

k={tr...

Gre

...

>...

System Level Model

Subcomponent Level Models Conservation Equations

Component Level Models

//Dynamic Mass Balance

M_x = transposex*M;

for i in 1:nspecies loop

derM_x[nspecies] = summdot_x[:, nspecies]

end if;

//Dynamic Energy Balance

U = M*h - p*Vt;

derU = sumqdot + sumheat.Q_s + sumheat.W_loss;

// Volume conservation

M[1] = d[1]*V[1];

LiBr Modelica component libraries built in collaboration with

SJTU

Modelica Modelling for HIL

3E4 3.5E4 4E4 4.5E4 5E4 5.5E4278

279

280

281

282e.y[11] chiller.Tchw out

3E4 3.5E4 4E4 4.5E4 5E4 5.5E4

302

304

306

308

e.y[12] chiller.Tcw out

Sensor to model validation

Use of Modelica for:

1. Controller certification (testing) in a

Hardware in the Loop environment

2. Control education

Modelica Modelling for Control Education

Pollock Theater est. 2010

298 stadium-seat auditorium

Sony 4K forever imaging

Meyer Sound offers Dolby 5.1 sound with

14 surround speakers.

2 Dual Kinoton16MM/35MM Film

Projectors

05/17 05/18 05/19 05/20 05/21 05/22 05/23

65

70

75

80

madx:NAE13/N2-1.277-AHU-T1-ZN2.SA1-T.Present Value

05/19 05/20

65

70

75

80

madx:NAE13/N2-1.277-AHU-T1-ZN2.SA1-T.Present Value


mix?

yATM?T

syst?

gd?

Su?

Re?

Zone

VMk?

VMk?

ATM?T

se?

C

se?

C

Zon?

star?

Occ?

peri?

SAv?

Zone

sen?

sen?

Occ?

peri?

Sup?

u

Tabl? Tabl? Tabl?

Tabl?

Tabl?

k?

k=?

Sc?

Tabl? F?

T?

0E0 1E5 2E5 3E5 4E5

68

69

70

71

72

73sensT_VAVTheater outT_VAVTheater


Objective

-Create a virtual theater to

illustrate benefits of alternative

control approaches

Approach

- Create a simple validated

Modelica model using the LBL

buildings library

Part III

Data-based analysis

Spatial-Frequency Analysis

0 20 40 60 80 100 120 140 160 18055

60

65

70

75

80

167

169

171

173

175

177

179

176

178

181

183

185

187

189

191

1931

1932

152

154

146

144

180

182

122

120

153

155

159

161

145

147

149

151

135

139

141

143

127

129

131

133

119

117

121

100

106

O1000

Sensor

Information

O10

Engineering

Information

With I. Mezic UCSB

Spatial Energy Visualization

Building Information + Installed Sensor Information + Wireless Mesh Information

Mathematics

Information

Spatial-Frequency Analysis

06/02 06/03 06/04 06/05

60

65

70

06/02 06/03 06/04 06/05

70

72

74

Outdoor air

temperature Indoor air

temperatures

Phase Delay Amplification or

Attenuation

Typical Sensor Trends

Challenge and Solution

numerous data outputs + numerous frequencies

too much data to process

•Control •Meetings / classes •Diurnal •Weekly •Seasonal

The challenge:

inter-dependent global output

+ spectral decompostion

minimal & definitive output

The solution:

Useful for analysis

and model validation

Decomposition methods

Previous work Principle Component Analysis PCA, Proper Orthogonal Decomposition POD,

etc

[Lam 2010]: PCA over long periods of time 29 years to study climate vs.

energy usage in a building

[Ruch 1993]: PCA to study before and after performance of building retrofits

[Du 2007]: Fault detection

[Reginato 2009]: Prediction of energy usage

……

Gaps / Drawbacks: Gaps: - Not used for model calibration Drawback of POD methods: - An energy based method, energy may not be best way to sort modes -Time dependencies are lost e.g. spectral content - Phase information is lost

Mathematical Preliminaries

The Koopman operator U is an infinite

dimensional operator that maps g to Ug

Begin with an arbitrary finite dimensional

nonlinear function/system/model f

The goal is to find the dynamical properties spectral content, orthonormal basis, etc. of the operator Ug

[Mezic 2005, Nonlinear Dynamics]

Model or Data

M is an arbitrary manifold

Actual sensed variables


To analyze building system data generated from model or sensors, consider a snapshot matrix:


If the original process is linear *sidenote

Analysis of Ug:

Note that is a Krylov sequence

There exists iterative methods e.g. the Arnoldi method for spectral analysis of a Krylov sequence

[Schmid 2009 J. Fluid Mech.]

Also,

Koopman Approach in other works

[Schmid 2009 J. Fluid Mech.]

[Susuki 2010 IEEE Trans. on Power Systems ] [Rowley 2009 J. Fluid Mech.]

The Koopman operator defined in the 1930s, recently been used for analysis of other dynamical systems

Fluid Flow

Power Grid Swing Instability

….building

system

examples

Example – Y2E2 Stanford*

Yang and Yamazaki

Environment and Energy Building

166,000 ft^2 15k m^2

Designed 50% below

ASHRAE 90.1.2004 Passive daylighting, heating, cooling

and ventilation; automatically

controlled operable windows,

sunshades and fins; radiant floor

heating; active chilled beam cooling;

heat recovery ventilators;

photovoltaics; recycled water

• Sensors:

~2400 data points, down

to the minute

* With Tobias Maile, Martin Fischer Stanford

Koopman Approach

0 20 40 60 80 100 120 140 160 18055

60

65

70

75

80

167

169

171

173

175

177

179

176

178

181

183

185

187

189

191

1931

1932

152

154

146

144

180

182

122

120

153

155

159

161

145

147

149

151

135

139

141

143

127

129

131

133

119

117

121

100

106

Original and complicated time domain data

…..Step 1

Koopman Approach

Diurnal Work-day 7-8 hrs

Investigate and choose freq. in Koopman spectrum

…..Step 2

Koopman Approach

Magnitude and phase of Koopman mode

quickly illustrates performance

Architectural floor plan, first floor sensors

…..Step 3

Example: Inefficient Control

Method quickly isolates sensor / control issues

Energy at

unexpected

frequencies

Cycling found in control system

System retuned to reduce cycling

Example – Model Tuning

Comparison between extensive EnergyPlus model and data

Data

Model

Spectrum Magnitude

Eisenhower [Simbuild 2010]


Comparison of magnitude and spectra between model / data looks good.

Phase information between outside conditions and building temperatures has some discrepency

Phase delay occurs due to occupancy / HVAC scheduling, thermal mass, etc.

Data

Model

Phase


To better capture phase, model was altered 1. Active beam induction ratio was changed from 50% to 75% recirculated

air

2. Infiltration values were increased

3. Natural ventilation was introduced between corridors and offices

Model

Phase – Before Tuning Phase – After Tuning

Hong Kong: Efficiency analysis*

No data

No data

One Island East – Westlands Rd. Hong

Kong

70 story sky-scraper

Data: 11/1/2009 – 11/15/2009

N

* With Walter Yuen, Hong Kong Poly. Univ.

Hong Kong: Efficiency analysis

Out-of-phase controller response

one heating, one cooling is usually

indicative of inefficient operation

210 220 230 240 250 260 270

21

22

23

24

25

26

Hours

1

2

3

Irregular lunch-time

scheduling in some loops

are leading to awkward

control response

Questions?

an integrated framework for parametric design using building … · 2020. 7. 2. · michael...

Documents