+ All Categories
Home > Documents > Jorge Ortiz. Metadata verification Scalable anomaly detection.

Jorge Ortiz. Metadata verification Scalable anomaly detection.

Date post: 11-Jan-2016
Category:
Upload: norman-moore
View: 220 times
Download: 0 times
Share this document with a friend
22
Metadata Verification and SBS Jorge Ortiz
Transcript
Page 1: Jorge Ortiz.  Metadata verification  Scalable anomaly detection.

Metadata Verification and SBS

Jorge Ortiz

Page 2: Jorge Ortiz.  Metadata verification  Scalable anomaly detection.

Outline

Metadata verification Scalable anomaly detection

Page 3: Jorge Ortiz.  Metadata verification  Scalable anomaly detection.

Chiller

Pump

Chiller

Pump

AHUSF EF

Vent Vent

Zone

Page 4: Jorge Ortiz.  Metadata verification  Scalable anomaly detection.

Chiller

Pump

Chiller

Pump

AHUSF EF

Vent Vent

Zone

System

Space

Page 5: Jorge Ortiz.  Metadata verification  Scalable anomaly detection.

Types of relationships

Geometric Placement, associations

Functional Temperature, pressure, flow, etc.

Semantic Electrical device taxonomy Ownership

Page 6: Jorge Ortiz.  Metadata verification  Scalable anomaly detection.

Metadata management pipeline

Current Our work

Page 7: Jorge Ortiz.  Metadata verification  Scalable anomaly detection.

Geometric Verification

Are the geometric (spatial) associations correct?

Are all the sensors with the same spatial grouping in the same location? Sensors can be moved or replaced Contractor mislabels point in BMS

How can the sensor data guide this process?

SODA4R520__ART

Page 8: Jorge Ortiz.  Metadata verification  Scalable anomaly detection.

Similar Trend in Data Streams

Sensor streams driven by same phenomena

Common trend ineffective at uncovering relationships

Page 9: Jorge Ortiz.  Metadata verification  Scalable anomaly detection.

No Discernible Correlation Pattern in Raw Traces

Each row/column is a location in the building Each location has

one or more sensors

Cell (i,j) is the average device pairwise correlation between sensors at locations i and j

Page 10: Jorge Ortiz.  Metadata verification  Scalable anomaly detection.

Empirical mode decomposition Approach used for finding underlying

data trends Algorithm for decomposing signals in

the time domain of non-stationary, non-linear signals Similar to FFT, PCA but yields

characteristic time and frequency scales Output “Intrinsic mode functions”

Combination of underlying signal in the same time scale

Page 11: Jorge Ortiz.  Metadata verification  Scalable anomaly detection.

Devices in the same room

Page 12: Jorge Ortiz.  Metadata verification  Scalable anomaly detection.

Devices is different rooms

Page 13: Jorge Ortiz.  Metadata verification  Scalable anomaly detection.

Broader validation

Compare the EHP to 674 other sensors:

 

EMD helps us to discriminate un/related sensors

**Suggests Geometric Verification is possible**

Page 14: Jorge Ortiz.  Metadata verification  Scalable anomaly detection.

Functional Verification

Mislabeled “type” information of a data stream

Fault detection Strip, bind, and search process

Page 15: Jorge Ortiz.  Metadata verification  Scalable anomaly detection.

Buildings Generate Lots of Data Difficult for building managers to know where

to start to look for problems Which devices? Locations? Patterns? Time interval?

Key Observation Devices are used simultaneous in the same way Typically usage times/patterns are tightly un/coupled

▪ Example:▪ Lights and HVAC during the day

Basic assumption Normal usage is efficient.

Pairwise correlation analysis of sensor traces Uncover usage relationships between devices

Page 16: Jorge Ortiz.  Metadata verification  Scalable anomaly detection.

Strip and Bind

Page 17: Jorge Ortiz.  Metadata verification  Scalable anomaly detection.

Searching for Outliers

Construct reference matrix for each time-reference interval

For new data points, compute l

Identifying outliers Median Absolute Deviation

p=4

, b=1.4826

Page 18: Jorge Ortiz.  Metadata verification  Scalable anomaly detection.

Results

High power usage Alarms corresponding to

electricity waste Lower power usage

Alarms representing abnormal low electricity consumption

Punctual Short increase/decrease in

electricity consumption Missing data

Possible sensor failure Other

unknown

Page 19: Jorge Ortiz.  Metadata verification  Scalable anomaly detection.

Alarms in Eng. Bldg 2 @Todai

AC On All Night Lights On All NightAC Not On DuringOffice Hours

Page 20: Jorge Ortiz.  Metadata verification  Scalable anomaly detection.

Alarms in Cory Hall

Possible Chiller dysfunction Change in power usage pattern

Simultaneous heating and cooling

Normal

18 days, 2500 kWh

Page 21: Jorge Ortiz.  Metadata verification  Scalable anomaly detection.

Impact

2 research papers in collaboration with U. of Tokyo Internet of Thing Workshop @IPSN 2012 IPSN 2013 (April)

Web tool that finds anomalies from data uploads Upcoming release

Page 22: Jorge Ortiz.  Metadata verification  Scalable anomaly detection.

Future Work

Value verification Model-based verification, model

validation Standard representation with

embedded confidence parameters for MPC


Recommended