David E. Culler Randy Katz Francesco Borrelli 1-11-2013 “A House Is a Machine for Living In.” - Le Corbusier.

Software Defined Buildings Pretreat

David E. CullerRandy KatzFrancesco Borrelli1-11-2013

“A House Is a Machine for Living In.” - Le Corbusier

Who Is Here - Introductions

UCB Software Defined Buildings Project UCB Center for the Built Environment UC California Institute for Energy and

Environment LBNL – EETD, Sustainability, … NREL Integral Group Intel, IBM, … Univ. of L’Aquila

Project Goals &Technology Transfer

3

UC Berkeley Project Team CollaboratorsSponsorsFriends

PeopleProject Status

Work in ProgressPrototype Technology

Early Access to TechnologyPromising Directions

Reality CheckFeedback

Agenda9:30 - 10:00 Coffee10:00 - 10:30 Introductions/Project Overview: David Culler, Randy Katz, Francesco Borrelli10:30 – 12:15 Platform

BOSS: software architecture for security and reliability -- Stephen Dawson-Haggerty

Building Application Stack - an initial runtime and API -- Andrew KrioukovPrivacy and Security -- Prashanth MohanOverview of MPC in Buildings -- Tony Kelman

12:15 – 1:15 Lunch1:15 – 2:15 Modeling, Learning, and Control

Berkeley Library for Optimization Modeling -- Sergey VichikPhysical and empirical modeling -- Jason TragerAnomaly detection and relationship inference -- Jorge OrtizLocalization -- Kaifei Chen

2:15 – 3:00 Infrastructure and ApplicationsMPC Lab HVAC system (µBuilding) -- Yudong MaDemand-controlled ventilation, and its grid potential -- Jay TanejaBuilding application targets -- Andrew Krioukov

3:00 – 3:30 Break/Demos3:30 – 4:00 Feedback and Discussion4:00 – 5:00 Beer/Wine Reception and Informal Discussion

Buildings …

Where we spend 90% of our lives Where we spend 70% of our electricity Where we spend 40% of our energy Where we spend 40% of our CO2

emissions Where we spend a lot of our $’s

And once they are built all we can do … is use their hard-wired capabilities, decorate, or “retrofit”

1/11/13SDB pretreat 5

How can we make them fundamentally more agile machines?

Programmable Separation of the hardware capabilities

(primitives) From the universe of potential behaviors

(applications) Allow them to be tailored to our desires

To the full extent of the underlying capabilities

Become a good citizen of a broader ecosystem

1/11/13SDB pretreat 6

BMS

Cyber Physical BuildingL

igh

t

Tra

ns

po

rt

Process Loads

Occupant Demand

Legacy Instrumentation & Control Interfaces

Pervasive Sensing

Activity/Usage

Streams

Local Controllers

Planning

VisualizationOccupant

SatisfactionMulti-Objective Model-Driven

Control

Building IntegratedOperating System

HV

AC

Ele

ctri

cal

Se

cu

rity

, F

au

lt,

An

om

aly

De

tec

t &

Ma

na

ge

me

nt

Control / Schedule

External

Human

-Build

ing

Inte

rface

BIM proxydrvrs

Energy Environment Outdoor EnvironmentPersonal Environment

Mapping

SoftZones

Physical Info Bus

privacy-pres. query

Elements of a Software Defined Building

1/11/13SDB pretreat 7

Physical Models

Empirical Models

App

sandbox

Pieces that we know we can do Building Application Programming Interface

cf., Andrew’s BAS Building Operation System & Services

Physical services and distributed device drivers Middle services: mapping, transactions, RAS Application services: baselining, ensemble, … cf., Stephen’s BOSS

Innovate in Model-Driven Predictive Control With interesting objectives: supply-following, cf,., Tony’s MPC

Rich Human-Building Interaction Location, personal and ambient devices, gestures, … cf.,

Kaifei Introduce meaningful security

Although a broad open research agenda in each

1/11/13SDB pretreat 8

Wide Open Challenges

Scale ~110 M buildings in US To make a difference everything has to be

automated after insertion of basic capability Heterogeneity

in Design, Implementation, Use, … Automated metadata acquisition & context Learning throughout lifecycle

Uncertainty In time, space, use, behavior, …

Empowerment and Balance Privacy, security, autonomy, control, opportunity

1/11/13SDB pretreat 9

Some Building Applications

Whole-Building Integrated Optimized Control Supply-following Utilization of Passives and Occupancy

In situ model building Prognostics, diagnostics, logistics

Personalized building interactions Cell phone as HBCI

Localization via WiFi, sensing, participation Free space gestures

Softzones, Microzones …

1/11/13SDB pretreat 10

Components of a BOSS

Physical Information Bus Historian Model Building Apparatus Control Application Sandbox Transaction Monitor Mapper Privacy Preserving Query Processing Personalized and Physicalized

Human-Building Interface

1/11/13SDB pretreat 11

Automated metadata ingestion and representation for buildings (Arka)

The problem: Large manual effort needed to construct building metadata in

order to run applications. Lots of Different Disconnected metadata sources :

BIM model, BMS software, CAD drawings, BACNet discovery, etc. Imagery, Occupant Interactions, …

Solution : Design adapters for ingest from these diverse information

sources. Automate rudimentary building metadata database formation Refine over the lifecycle

Maintain a standard representation of a building against which applications can be written.

Have mechanisms for conflict resolution of ingested information

Explicitly represent uncertainty in the building representation

CAD files

BIM Model

Google Sketchup Model

BMS web tier

Adapter 1

Adapter 2

Adapter 3

Adapter 4

Adapter n

BACNet points

HVAC app

Modeling

software

Version Control

Renderer

Type checkerConstra

int Propaga

torManual Input

Lighting App

INFORMATION SOURCES

BUILDING REPRESENTATION :

gbXML+

Representation of: (1) Temporal

Uncertainty (changes during building lifecycle)

(2) Data Ingestion Uncertainty (inomplete/incorrect information sources)

(3) Spatial uncertainty [e.g exact schedules of a building within a larger campus]

UNCERTAINTY REDUCER

BUILDING APPS

CO

NFLIC

T R

ES

OLU

TIO

N

Data Ingestion Example: ALC web portal adapter

INPUT SOURCE : ALC BMS web page of

building “DOE Annex”

<Airloop systemType="VariableAirVolume"> <AirLoopEquipment equipmentType="VAVBox” id=“doe_vav_b-4-01”> <ShellGeometry> <ClosedShell> <PolyLoop> <CartesianPoint> <Coordinate> 13 </Coordinate> <Coordinate> 51 </Coordinate> <Coordinate flag="AddedByIngestor"> 10.0 </Coordinate> ………</AirLoop>

REPRESENTATION

bacnet ID

Ingestion Uncertaintyof the z-coordinateADAPTER

CAD files

BIM Model

Google Sketchup Model

BMS web tier

Adapter 1

Adapter 2

Adapter 3

Adapter 4

Adapter n

BACNet points

HVAC app

Modeling

software

Version Control

Renderer

Type checkerConstra

int Propaga

torManual Input

Lighting App

INFORMATION SOURCES

BUILDING REPRESENTATION :

gbXML+

Representation of: (1) Temporal

Uncertainty (changes during building lifecycle)

(2) Data Ingestion Uncertainty (inomplete/incorrect information sources)

(3) Spatial uncertainty [e.g exact schedules of a building within a larger campus]

UNCERTAINTY REDUCER

BUILDING APPS

CO

NFLIC

T R

ES

OLU

TIO

N

Can we Make Buildings Greener?

Environment

Humans Building

PredictiveController

Predictions on Building Dynamics, Weather, Occupancy, Comfort

Model Predictive Control / Learning

Average energy consumption reduction of 60-85%over DDC mode levels.

Source: “Model Predictive Control for Mid-Size Commercial Building HVAC.” Experimental work done by Dr. Borrelli group in conjunction with UTC Research Center and UC Berkeley.Published February 2012. US Army Corp of Engineers, Champaign, IL

Basic Idea

Avoid Region

Two Combined Effects : Anticipation and Coordination

y(k + 1) = y(k) + bww(k) + buu(k); y(k) 2 Y(k)

At step t decide on u(t) based on prediction on w(t),..., w(t+N), Y(t),…,Y(t+N)

human, environment constraints

Avoid Region

Avoid Region

y(k)

time

control action

Advantages:

• Predictive• Systematic: no if-then-else and extensive trial and error tuning• Multivariable, Model Based• Guarantees: Performance and Constraint satisfaction• Large success in the process industry• Flexible/ Easy to Integrate

P re d ic te d o u tp u ts

M a n ip u la te d ( t+ k )uIn p u ts

t t+ 1 t+m t+ p

fu tu rep a s t

t+ 1 t+ 2 t+ 1+m t+ 1+ p

Model Predictive Control (MPC)

Challenges

www.mpc.berkeley.edu

• “System” Knowledge – Right Model Abstraction• Predictions are uncertain• Large-scale• Scalability• Limited computational resources• Certification

“Better” Strategy

www.mpc.berkeley.edu

Autonomous Driving Volvo Experiments 2012

2012 IEEE Control System Technology Award

Discussion

1/11/13SDB pretreat

24