Model Based Event Detection in Sensor Networks Jayant Gupchup, Andreas Terzis, Randal Burns, Alex Szalay.

Slides:



Advertisements
Similar presentations
Applications of one-class classification
Advertisements

PCA for analysis of complex multivariate data. Interpretation of large data tables by PCA In industry, research and finance the amount of data is often.
Air Force Technical Applications Center 1 Subspace Based Three- Component Array Processing Gregory Wagner Nuclear Treaty Monitoring Geophysics Division.
Sensor-Based Abnormal Human-Activity Detection Authors: Jie Yin, Qiang Yang, and Jeffrey Junfeng Pan Presenter: Raghu Rangan.
Life Under Your Feet Johns Hopkins University Computer Science Earth and Planetary Sciences Physics and Astronomy
The Evolution of Spatial Outlier Detection Algorithms - An Analysis of Design CSci 8715 Spatial Databases Ryan Stello Kriti Mehra.
1er. Escuela Red ProTIC - Tandil, de Abril, 2006 Principal component analysis (PCA) is a technique that is useful for the compression and classification.
Image Enhancement Digital Image Processing Instructor: Dr. Cheng-Chien LiuCheng-Chien Liu Department of Earth Sciences National Cheng Kung University Last.
1 Vertically Integrated Seismic Analysis Stuart Russell Computer Science Division, UC Berkeley Nimar Arora, Erik Sudderth, Nick Hay.
“Real-time” Transient Detection Algorithms Dr. Kang Hyeun Ji, Thomas Herring MIT.
Probabilistic Aggregation in Distributed Networks Ling Huang, Ben Zhao, Anthony Joseph and John Kubiatowicz {hling, ravenben, adj,
Principal Component Analysis
Dimension reduction : PCA and Clustering Agnieszka S. Juncker Slides: Christopher Workman and Agnieszka S. Juncker Center for Biological Sequence Analysis.
Distributed Regression: an Efficient Framework for Modeling Sensor Network Data Carlos Guestrin Peter Bodik Romain Thibaux Mark Paskin Samuel Madden.
Microsoft E-Science Data Storage Model for Environmental Monitoring Wireless Sensor Networks Jayant Gupchup †, R.
UNIVERSITY OF SOUTH CAROLINA Department of Computer Science and Engineering On-line Alert Systems for Production Plants A Conflict Based Approach.
Multi-Scale Analysis for Network Traffic Prediction and Anomaly Detection Ling Huang Joint work with Anthony Joseph and Nina Taft January, 2005.
Dimension reduction : PCA and Clustering Christopher Workman Center for Biological Sequence Analysis DTU.
PSYC512: Research Methods PSYC512: Research Methods Lecture 15 Brian P. Dyre University of Idaho.
Laurent Itti: CS599 – Computational Architectures in Biological Vision, USC Lecture 7: Coding and Representation 1 Computational Architectures in.
Energy-efficient Self-adapting Online Linear Forecasting for Wireless Sensor Network Applications Jai-Jin Lim and Kang G. Shin Real-Time Computing Laboratory,
PSYC512: Research Methods PSYC512: Research Methods Lecture 14 Brian P. Dyre University of Idaho.
1 Activity and Motion Detection in Videos Longin Jan Latecki and Roland Miezianko, Temple University Dragoljub Pokrajac, Delaware State University Dover,
World Renewable Energy Forum May 15-17, 2012 Dr. James Hall.
Tomer Sagi and Avigdor Gal Technion - Israel Institute of Technology Non-binary Evaluation for Schema Matching ER 2012 October 2012, Florence.
INTEGRATION OF ARTIFICIAL INTELLIGENCE [AI] SYSTEMS FOR NUCLEAR POWER PLANT SURVEILLANCE & DIAGNOSTICS.
U NIVERSITY OF M ASSACHUSETTS A MHERST Department of Computer Science 2011 Predicting Solar Generation from Weather Forecasts Using Machine Learning Navin.
A Generalized Likelihood Ratio Approach to the Detection of Jumps in Linear Systems Min Luo.
IN23A-1072: Life Under Your Feet: A Wireless Soil Ecology Sensor Network K. Szlavecz 1, A. Terzis 1, R. Musaloiu 1, A. Szalay 1, J. Gupchup 1, C.-J. Liang.
IDIES Temporal Integrity Challenges in Long-term Environmental Monitoring Sensor Networks. Jayant Gupchup † Alex.
Building and End-to-end System for Long Term Soil Monitoring Katalin Szlávecz, Alex Szalay, Andreas Terzis, Razvan Musaloiu-E., Sam Small, Josh Cogan,
BACKGROUND LEARNING AND LETTER DETECTION USING TEXTURE WITH PRINCIPAL COMPONENT ANALYSIS (PCA) CIS 601 PROJECT SUMIT BASU FALL 2004.
Life Under Your Feet. Sensor Network Design Philosophies Use low cost components No access to line power –Deployed in remote locations Radio is the.
Towards a Model-Based Data Collection Framework for Environmental Monitoring Networks Research Proposal Jayant Gupchup Department of Computer Science,
Time Series Data Analysis - I Yaji Sripada. Dept. of Computing Science, University of Aberdeen2 In this lecture you learn What are Time Series? How to.
Microsoft E-Science Data Storage Model for Environmental Monitoring Wireless Sensor Networks Jayant Gupchup †, R.
Phase Congruency Detects Corners and Edges Peter Kovesi School of Computer Science & Software Engineering The University of Western Australia.
Latest results in verification over Poland Katarzyna Starosta, Joanna Linkowska Institute of Meteorology and Water Management, Warsaw 9th COSMO General.
Gap-filling and Fault-detection for the life under your feet dataset.
K. Kolomvatsos 1, C. Anagnostopoulos 2, and S. Hadjiefthymiades 1 An Efficient Environmental Monitoring System adopting Data Fusion, Prediction & Fuzzy.
Jayant Gupchup Phoenix, EWSN 2010 Phoenix: An Epidemic Approach to Time Reconstruction Jayant Gupchup †, Douglas Carlson †, Răzvan Musăloiu-E. †,*, Alex.
CCN COMPLEX COMPUTING NETWORKS1 This research has been supported in part by European Commission FP6 IYTE-Wireless Project (Contract No: )
ELECTIONEL ECTI ON ELECTION: Energy-efficient and Low- latEncy sCheduling Technique for wIreless sensOr Networks Shamim Begum, Shao-Cheng Wang, Bhaskar.
Secure In-Network Aggregation for Wireless Sensor Networks
Principal Components Analysis. Principal Components Analysis (PCA) A multivariate technique with the central aim of reducing the dimensionality of a multivariate.
J.-Y. Yang, J.-S. Wang and Y.-P. Chena, Using acceleration measurements for activity recognition: An effective learning algorithm for constructing neural.
EE4-62 MLCV Lecture Face Recognition – Subspace/Manifold Learning Tae-Kyun Kim 1 EE4-62 MLCV.
Christina Bonfanti University of Miami- RSMAS MPO 524.
Syed Hassan Ahmed Syed Hassan Ahmed, Safdar H. Bouk, Nadeem Javaid, and Iwao Sasase RIU Islamabad. IMNIC’12, RIU Islamabad.
Evaluating temporal and spatial O 3 and PM 2.5 patterns simulated during an annual CMAQ application over the continental U.S. Evaluating temporal and spatial.
EE515/IS523: Security 101: Think Like an Adversary Evading Anomarly Detection through Variance Injection Attacks on PCA Benjamin I.P. Rubinstein, Blaine.
Beyond Spectral and Spatial data: Exploring other domains of information: 3 GEOG3010 Remote Sensing and Image Processing Lewis RSU.
Combined Human, Antenna Orientation in Elevation Direction and Ground Effect on RSSI in Wireless Sensor Networks Syed Hassan Ahmed, Safdar H. Bouk, Nadeem.
Experiences and Challenges in Campaign Style Deployments using Wireless Sensor Networks Jayant Gupchup †, Scott Pitz *, Douglas Carlson †, Chih-Han Chang.
Locations. Soil Temperature Dataset Observations Data is – Correlated in time and space – Evolving over time (seasons) – Gappy (Due to failures) – Faulty.
LIGO-G Z r statistics for time-domain cross correlation on burst candidate events Laura Cadonati LIGO-MIT LSC collaboration meeting, LLO march.
Similarity Measurement and Detection of Video Sequences Chu-Hong HOI Supervisor: Prof. Michael R. LYU Marker: Prof. Yiu Sang MOON 25 April, 2003 Dept.
How Dirty is your Data : The Duality between detecting Events and Faults J. Gupchup A. Terzis R. Burns A. Szalay Department of Computer Science Johns Hopkins.
Forecasting. Model with indicator variables The choice of a forecasting technique depends on the components identified in the time series. The techniques.
Enabling Real Time Alerting through streaming pattern discovery Chengyang Zhang Computer Science Department University of North Texas 11/21/2016 CRI Group.
Experience Report: System Log Analysis for Anomaly Detection
Antecedent Environments Conducive to the Production of Extreme Temperature and Precipitation Events in the United States Andrew C. Winters, Daniel Keyser,
Meng Lu and Edzer Pebesma
Long-range capacitive sensors for indoor person location
Quality Control at a Local Brewery
SLOPE: A MATLAB Revival
IX International Workshop ACAT
Center for Big Data Analysis
K. Kolomvatsos1, C. Anagnostopoulos2, and S. Hadjiefthymiades1
Principal Component Analysis
Presentation transcript:

Model Based Event Detection in Sensor Networks Jayant Gupchup, Andreas Terzis, Randal Burns, Alex Szalay

The Johns Hopkins University2 Outline Motivation Data and Model Experiments and Results Discussion

The Johns Hopkins University3 Event starts Detect Event Increase Sampling Frequency/Trigger Alarms Event ends Return to steady behavior Data Sampling in WSNs -Most environmental monitoring networks today sample at fixed frequencies -The failure of fixed frequency sampling -High Frequency: Generates large volumes of measurements -Low Frequency: Misses temporal transients -Solution: Adaptive Sampling based on the ability to detect events of interest -Benefits -Less but more “interesting” data -Conserve energy -Trigger alarms -Event-based querying in the back-end database

The Johns Hopkins University4 Rain Event Non-Event Days Sample Event

The Johns Hopkins University5 Solution Outline Model observed phenomena using Principal Component Analysis (PCA) Project original measurements on to a feature space –Benefit: reduces dimensionality Look for measurements deviating from average/expected behavior in the feature space

The Johns Hopkins University6 First Principal Component Variable #1 Variable #2 X : Points in the original space O : Projection on PC1 Principal Component Analysis PCA : Finds axes of maximum variance in the collected data Reduces original dimensionality –Example: 2 variables  1 variable

The Johns Hopkins University7 Motivation for Using PCA Typical day: “Fits model well”Event day: “Large residuals”

The Johns Hopkins University8 Specific Application LifeUnderYourFeet: Environmental Monitoring network for soil moisture Deployment details –10 MICAz Sensors Air Temperature (AT) Soil Temperature (ST) Soil Moisture Light intensity –Deployed for a period of a year Goal: Detect significant rain events

The Johns Hopkins University9 Why not Soil Moisture ? Reaction to event

The Johns Hopkins University10 Air Temp vs. Soil Temp Notice the phase lag for Soil Temperature

The Johns Hopkins University11 Data Preparation Build model for Air and Soil temperature AT 1_1 AT 1_2 …. … AT 1_144 AT 2_1 AT 2_2 …. … AT 2_144.. …. …... …. AT 10_1 AT 10_2 …. … AT 10_144.. …. …... …... …. t=10 t=20 … t= day, 10 sensors Size of matrix: [( # of days x 10 )  144 ]

The Johns Hopkins University12 Number of PCA basis required

The Johns Hopkins University13 PCA Bases (AT & ST) Eigenvector1 Is the Diurnal cycle similarity eigenvector1 for ST & eigenvector2 for AT

The Johns Hopkins University14 Event Detection Methods 1.Basic Method –Projections on the first principal component for AT 2.Highpass Method –Removes seasonal drift by looking at sharp changes in the local neighborhood 3.Delta method –Uses the inertia of the soil and seasonal drift

The Johns Hopkins University15 Method 1 : Basic Method Considers only Air Temperature First Basis Vector covers 55% of variation in the data AT 1_1 AT 1_2 …. … AT 1_144 AT 2_1 AT 2_2 …. … AT 2_144.. …. …. AT 10_1 AT 10_2 …. … AT 10_144 V 1_1 V 1_2. V 1_144 e 1_1 e 2_1. e 10_1 X = Average E1E1 E2E2 …….. ……………..E n-1 EnEn Day 1Day 2Day n 10 sensors First Basis Vector (PC1) 1 day Apply threshold on E 1 series Tag values below the threshold as events

The Johns Hopkins University16 Method 2 : Highpass Method Again, considers only Air Temperature Apply highpass filter on E 1 series  S 1 series Highpass filter detects sharp changes by focusing on a limited time window  removes seasonal drift Apply threshold on S 1 series –Tag values below the threshold as events

The Johns Hopkins University17 Method 3 : Delta Method Considers Air Temperature (AT) and Soil Temperature (ST) Create E 1 series for AT and ST Apply Highpass filter on E 1,AT & E 1,ST  S 1,AT & S 1,ST Compute Delta = S 1,AT - S 1,ST for all days Set a threshold on the Delta series

The Johns Hopkins University18 Evaluation Data Set –Test Period : 225 days between September, 2005 – July, 2006 –48 major rain events occurred during this period Reported by the BWI weather station Evaluation metrics –Precision (true positives) –Recall –Number of false negatives

The Johns Hopkins University19 Results MethodPrecisionRecallFalse Negatives Basic52.459%64%18 Highpass51.28%80%10 Delta54.79%85.106%7 Method shortcomings -Does not consider seasonal drift (Basic) -Does not use the inertia information of the soil (Basic, Highpass)

The Johns Hopkins University20 Event detection for 12/13/2005 – 01/02/2006 Due to the inertia of the soil, ‘Delta method’ shows sharper negative peaks for event days.

The Johns Hopkins University21 Future work Implement “Online event detection” –Compute Basis vectors from historic data –Load the ‘basis vectors’ and ‘threshold’ values on the motes Detect localized events by forming clusters of motes with similar eigen-coefficients Apply technique for faulty sensor detection Consider variants of PCA (Gappy-PCA, online-PCA)

The Johns Hopkins University22 Acknowledgements Ching-Wa Yip 1 - PCA C# library and Discussions. Katalin Szlavecz 2 & Razvan Musaloui-E 3 –Domain expertise and data collection. Jim Gray 4 & Stuart Ozer 4 –Online database 1 : JHU, Dept of Physics & Astronomy 2 : JHU, Dept of Earth and Planetary science 3 : JHU, Dept of Computer Science. 4 : Microsoft Research

The Johns Hopkins University23 Future work Online event detection on the motes Apply this method for faulty sensor detection Detect localized events by forming clusters of motes with similar eigencoefficients. Consider incomplete days using Gappy-PCA. Explore incremental & robust PCA techniques.

The Johns Hopkins University24 Training Set (Air Temp) Seasons exhibit “Diurnal Cycles” around their daily mean (DC component) Construct Zero-Mean Vectors for each Sensor i for each day (remove DC Component) Remove outliers using a simple median filter to build the training set X