Brett amidan Jim Follum Kimberly Freeman Jeff Dagle Pacific Northwest National Laboratory October 6 2015 1 CIGRE US National Committee 2015 Grid of the Future Symposium Big Picture Objective ID: 733037
Download Presentation The PPT/PDF document "Baselining PMU Data to Find Patterns and..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Baselining PMU Data to Find Patterns and Anomalies
Brett amidanJim FollumKimberly FreemanJeff Dagle
Pacific Northwest National Laboratory
October 6, 2015
1
CIGRE US National Committee 2015 Grid of the Future SymposiumSlide2
“Big Picture” Objective
Power grid related data (PMUs, State Estimators, Load, etc)
Analytical Tool that provides:
Real time analytics, monitoring the state of the grid
Capability to look at historical trends and events
Reliable predictions about the forthcoming state of the grid
2
October 6, 2015Slide3
Pre-Processing StepsRead raw PMU data
Develop and then use data quality filters to clean poor quality data
October 6, 2015
b.amidan@pnnl.gov
3
Remove bad data
1 day of 60 Hz PMU data (54 PMUs) = 26 GB
FrequencySlide4
Feature Extraction (Data Signatures)Regression fits through the data calculate estimates of value
, slope, curvature (acceleration), and noise
.Can be calculated in the presence of missing or data quality flagged values.
Summaries of these features are used in the analyses.
October 6, 2015b.amidan@pnnl.gov
4Slide5
Baselining Grid BehaviorUnivariate ApproachCreate a baseline of typical behavior for each individual variableDetermine abnormal behavior based on the baseline
Multivariate ApproachCreate a baseline across many (hundreds or even thousands) of variablesRelationship between variables is considered when determining abnormal behaviorOctober 6, 2015
5
StaticBaselining LimitsSlide6
October 6, 20156
Univariate Baselining Example
Model
Predicted Phase Angle PairValue at MidnightHours 0-23Day of Week
Date / Time Model – Time Series Based ModelPhase Angle Difference
Initial Training Period
Actual Value
Dynamic
Baselining
Limits
(Calculated Daily)Slide7
Multivariate BaseliningBaseline captures what normal behavior is expected to beGroup
similar behavior Time periods that group together indicate normal grid behaviorVariables that group together indicate highly correlated variables and may be candidates for feature reductionIdentify data that does not belong with the normal behavior
Time period contains data that is unusual (possible abnormal grid behavior)
Variable is unlike other variables, or something has happened to indicate a behavioral change in the variableOctober 6, 20157Slide8
October 6, 20158Creating a Baseline –
Unsupervised Learning
Baselining Learning Algorithm
ModelTraining Data:Historical PMU Data
Real Time PMU Data
Class 1
Class 2
Class 3
Class 4
Class 5Slide9
October 6, 2015
9
Identifying Data Driven Atypical Events
Using
multivariate
statistical techniques to establish baselines of typical behavior, atypical moments in time can be discovered and the variables responsible can be identified.Slide10
Atypicality DetectionLightning Related Anomaly
October 6, 2015
10
Atypicality Score
Substation A
Substation BSlide11
Atypicality DetectionEquipment Failure Related Anomaly
October 6, 2015
b.amidan@pnnl.gov
11
Atypicality Score
Other PMUs behaved similarlySlide12
Phase Angle Pairs Clustering
October 6, 2015
b.amidan@pnnl.gov
12
Proximity on tree indicates similarity
Unsupervised learning (clustering) used to determine which variables are most similar during
Time Period A
.Slide13
Phase Angle Pairs Clustering
October 6, 2015
b.amidan@pnnl.gov
13
Time Period B (two months later)
Phase Angle Pair #2 is no longer like Pair #1. Why?Slide14
October 6, 201514Supervised
Learning
Baselining Learning Algorithm
Predictive ModelTraining Data:Historical PMU Data
Real Time PMU Data
Class 1
Class 2
Class 3
Class 4
Class 5
Labels
Normal
Voltage Drop
Surge
Maintenance
Weather
Weather
Normal
Voltage Drop
Surge
MaintenanceSlide15
October 6, 201515
Understanding Precursors to Inform Prediction Models
Known event
Precursor activity
Precursor Features (Signature)Inform Machine- Learning Model
Create Classification to identify future precursorsSlide16
October 6, 201516
Future Step – Using Supervised Learning to Predict Current State
Extract Signature
Classification Based Prediction Model (Trained from Historical Data)
Event 1Normal 4Precursor 3
Event 7
0.75
0.15
0.05
0.04
Likelihood
Possible Patterns
NOTE: Only events and precursors with distinct data characteristics will be identifiable Slide17
ConclusionsData driven anomalies can be identified using multivariate analyses techniques. Some of these anomalies correspond to actual events, but some do not.Understanding precursors can inform prediction models, allowing for probability based predictions of the near-term future grid behavior.
October 6, 2015
17