monitoring and modeling with streamfs

24
Monitoring and Modeling with StreamFS Jorge Ortiz University of California, Berkeley

Upload: padma

Post on 24-Feb-2016

32 views

Category:

Documents


0 download

DESCRIPTION

Monitoring and Modeling with StreamFS. Jorge Ortiz University of California, Berkeley. Building energy consumption highly fragmented. Building Management System captures Heat/cooling and ventilation Lighting systems Miscellaneous electrical loads Weather data, price, etc. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Monitoring and Modeling with  StreamFS

Monitoring and Modelingwith StreamFS

Jorge OrtizUniversity of California, Berkeley

Page 2: Monitoring and Modeling with  StreamFS

2 /23

• Building Management System captures Heat/cooling and ventilation

• Lighting systems• Miscellaneous electrical loads• Weather data, price, etc.• Integration is key

Building energy consumption highly fragmented

HVAC: 31.4%

+

01/13/11

Page 3: Monitoring and Modeling with  StreamFS

3 /23

• Fault detection study[Schein2005]

• Fault: Simultaneous heating and cooling– Controllers on separate schedules

Why integrate?

Heating coil valvePosition varies

Outside-air mixerPosition varies

Cooling coil remains off

01/13/11

Page 4: Monitoring and Modeling with  StreamFS

4 /23

Integration helps deduce activity

• Human-activity classification– Electrical activity [Patel2007]– HVAC air pressure [Patel2008]– Water usage[Froehlich2008]– IP traffic and circuit-level activity [Kim2010]

• SmartThermostat[Lu2010]– Combines motion sensors and contact switches to

reduce HVAC energy consumption by 28%

01/13/11

Page 5: Monitoring and Modeling with  StreamFS

5 /23

Integration With CurrentSystems is Hard

01/13/11

Page 6: Monitoring and Modeling with  StreamFS

6 /23

Commercial BMS Architecture

Field Level

Routing/Controllers

Applications

01/13/11

Page 7: Monitoring and Modeling with  StreamFS

7 /23

Problems with BMS’s

• Not designed to collect all the data– Memory limit at control layer, application layer– Most information is lost through sense-point

“bundling” (averaging)• Burden on operator to manage

– Must decide which signals to “trend/unbundle”, monitor (set trigger)

– Leads to missing data in aggregate reports• Multi-signal fault detection done by human

operator01/13/11

Page 8: Monitoring and Modeling with  StreamFS

/23

The world is a nasty place

• State-of-art not designed for data collection• 30% of sensors are broken[BEMS2000]

– Mixed air reading errors +2.8 Celsius increases cooling energy consumption by 60% [Kao1983]

– Mixed are reading errors -2.8 Celsius increases heating energy consumption by 30% [Kao1983]

• Sensor data has fundamental problems– Data missing, variable production rate, calibration

necessary, multidimensional, etc [Balazinska2007]

8 01/13/11

Page 9: Monitoring and Modeling with  StreamFS

/23

What do we want to achieve?

• Generality– Supports the integration of many input/output sources

• Ease of use– Add/remove input sources, add/remove output targets– Querying/Cleaning/Sharing the data– Use the metadata to make more informed queries

• Organizing principle: Everything looks like a distributed file system– Hierarchy restricts data access through naming– Useful for accessing data according to semantic,

categorical, or physical placement9 01/13/11

Page 10: Monitoring and Modeling with  StreamFS

10 /23

Building multiview integration

Electrical Load Tree

Environment and Activity

Climate plant

/SodaHall

/hvac

/CT /Chiller

/loadtree

/panel /xform

/spaces

/floor3/floor4

01/13/11

Page 11: Monitoring and Modeling with  StreamFS

11 /23

Organizing the metadata

/SodaHall

/inventory

{“desc”:”inventory inside SDH”“timestamp”: …}

{“desc”:”Lamp”“timestamp”: …}

{“desc”:”Phone”“timestamp”: …}

{“desc”:”Outlet”“timestamp”: …}

{“desc”:”Acme”“timestamp”: …} r-node

s-node

/hvac

/CT /Chiller

/vent

/loadtree

/panel /xform

/outlet

/spaces

/floor3/floor4

/power/mote123

01/13/11

Page 12: Monitoring and Modeling with  StreamFS

12 /23

Data collection and querying

/inventory/mote123

DB

PID1

PID1

PID2

PID2

PID3

PID3

PID4

PID4

Tim

e

GET?query=true&ts_timestamp=gt:now-100,ls=now

GET/SDH/spaces/*?query=true&props_metertype=powermeter

{“metertype”:”powermeter”,“desc”:”Electric power meter”,“timestamp”: 1290500046}

/power /temp/hum/par

01/13/11

Page 13: Monitoring and Modeling with  StreamFS

13 /23

Data representation layer

• Narrow-waist for data representation– Simple Metering and Actuation Profile (sMAP)

sMAP

Electrical

Weather

GeographicalWater

EnvironmentalStructural Actuator

Occupancy

Phys

ical

In

form

ation

/ # list resource under URI root [GET] /data # list sense points under resource data [GET] / [sense_point] # select a sense points [GET] /meter # meters provide this service [GET] / [channel] # a particular channel [GET] /reading # meter reading [GET] /format # calibration and units [GET/POST] /parameter # sampling parameter [GET/POST] /profile # history of readings [GET] /report # create and query periodic reports [GET/POST]

01/13/11

Page 14: Monitoring and Modeling with  StreamFS

14 /23

RESTful + JSONInterface

{

"operation":"create_publisher","resourceName”:”power"

}

PUT http://is4server.com/is4/devices/mote123/

{”pubid":"550e8400",

}

REPLY: 201 Created

PUT http://is4server.com/is4/devices/mote123/power

{"Reading": 120,

}

POST http://is4server.com/is4/devices/mote123/power?pubid=550e8400

{“desc”:”Temperature mote”,"Reading": 120,“timestamp”: 1290500046}

GET http://is4server.com/is4/devices/mote123/power

01/13/11

Page 15: Monitoring and Modeling with  StreamFS

15 /23

Sharing real-time feeds

POST http://is4server.com/sub

{"streams":[550e8400],"url":"http://128.32.37.21:8011/

sub.php"}

{”subid":"41d4",

}

http://is4server.com/sub/41d4

REPLY: 201 Created

mote123/power

price

BMSZigbee

StreamFS

http://128.32.37.21:8011/sub.php

POST

550e8400

41d4

01/13/11

Page 16: Monitoring and Modeling with  StreamFS

/23

Standard distillationelements

• Provide regression, interpolation, extrapolation functions over space and time values

• Provide join and filter functions

• Related work: MauveDB[Deshpande2006] x

y

User

x

y

Consistent uniform view

Apply regression;Compute “temp” at grid points

1601/13/11

Page 17: Monitoring and Modeling with  StreamFS

17 /23

Data cleaning and distillation staging

InterpolateF1(x)

ExtrapolateF2(x)

/models /inventory/mote123

/power /current/interp /filter

/inventory/mote123/power | /models/interp | /models/filter

| http://128.32.37.21:8011/sub.php

/proc_chains/983hfq

Java/Javascript

01/13/11

Page 18: Monitoring and Modeling with  StreamFS

/23

Model resource example

18

{ winsize:10, materialize:true, timeout:2000, func:function(istream){ var ostream = new Object(); var sum =0; for(i=0; i<istream.length; i++){ var data = istream[i]; sum += data.Reading; }

ostream.points = istream.length; ostream.avg = sum/istream.length; return ostream; }}

01/13/11

Page 19: Monitoring and Modeling with  StreamFS

/23

Putting it all together (1)

19

/floor4

/room410

/therm/mote01

GET/…/floor4/room410/*/*?query=true&prop_type=temp

/temp /temp

InterpolateF1(x)

InterpolateF1(x)

/…/floor4/room410/room410/therm/temp

/…/floor4/room410/room410/mote01/temp

/…/floor4/room410/room410/mote01/temp?query=true&ts_timestamp=lte:t1,gte:t7

/…/floor4/room410/room410/therm/temp?query=true&ts_timestamp=lte:t1,lte:t7

01/13/11

Page 20: Monitoring and Modeling with  StreamFS

/23

Putting it all together (2)

20

/…/floor4/room410/*/*?query=true&type=device| /models/ts_getall?ts_timestamp=lte:t1,gte:t7

| /models/interp_all?attr=timestamp&unit=1

| /models/join?attr=timestamp

| http://viewer.com/viewer.phpt1

t2

t3

t5

t4

t6

t7

mot

e01/

tem

p

ther

m/t

emp In-time pipe-chainContinuous pipe-chain

01/13/11

Page 21: Monitoring and Modeling with  StreamFS

/23

Current Status

• Live instance on is4server.com– 2795 Incoming data streams– ~600 Kbps incoming data rate

• Releases available on the site– Version 2.0 with modeling to be released soon

• Documentation and tutorial available on is4server.com

• Used in various applications– Electrical load tree viewer, metadata graph viewer, SDH

Energy Audit application

21 01/13/11

Page 22: Monitoring and Modeling with  StreamFS

/23

Future work

• Inter-system cross-signal correlation– Build usage/fault signatures to detect inefficient

energy-use detection• How do we expand the use of StreamFS

beyond buildings?• Energy analytics

– Building blocks available, how do we use it?

01/13/11 22

Page 23: Monitoring and Modeling with  StreamFS

23 /23

Feedback/Questions?

01/13/11

Jorge Ortiz <[email protected]>

Page 24: Monitoring and Modeling with  StreamFS

/23 11/30/10 24