discovery through situational awareness - naspi€¦ · discovery through situational awareness....

19
Discovery Through Situational Awareness BRETT AMIDAN JIM FOLLUM CARLOS ORTIZ MARRERO Pacific Northwest National Laboratory March 29, 2017 1 NASPI – Session 3: Data and Data Quality March 2017

Upload: others

Post on 01-Jun-2020

9 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Discovery Through Situational Awareness - NASPI€¦ · Discovery Through Situational Awareness. Project Goal. Create a tool that applies statistical and machine-learning algorithms

Discovery Through Situational Awareness

BRETT AMIDANJIM FOLLUMCARLOS ORTIZ MARREROPacific Northwest National Laboratory

March 29, 2017

1

NASPI – Session 3: Data and Data QualityMarch 2017

Page 2: Discovery Through Situational Awareness - NASPI€¦ · Discovery Through Situational Awareness. Project Goal. Create a tool that applies statistical and machine-learning algorithms

DOE GMLC Award –Discovery Through Situational Awareness

Project GoalCreate a tool that applies statistical and machine-learning algorithms in context of big data analytics to investigate and implement anomaly and event detection algorithms in near real-time

Current FocusWorking with the Eastern InterconnectInitial focus on phase angle pair analysesProvide the EI partners with a frequent (i.e. daily or weekly) report of the findings

March 29, 2017 2

Page 3: Discovery Through Situational Awareness - NASPI€¦ · Discovery Through Situational Awareness. Project Goal. Create a tool that applies statistical and machine-learning algorithms

NASPI Contribution

EATT (Engineering Analysis Task Team) White Paper

Data Mining Techniques and Tools for Synchrophasor Data

March 29, 2017 3

A high level view of data mining –• What it is• Why it is important• How it -

1) has been used in other industries, 2) has been used in the power industry, and 3) may be used in the future

Page 4: Discovery Through Situational Awareness - NASPI€¦ · Discovery Through Situational Awareness. Project Goal. Create a tool that applies statistical and machine-learning algorithms

“Big Picture” Vision

Power grid related data (PMUs, State Estimators, SCADA, etc)

Data Driven 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 4March 29, 2017

Other data (i.e. weather [actual and forecasts], social media)

Page 5: Discovery Through Situational Awareness - NASPI€¦ · Discovery Through Situational Awareness. Project Goal. Create a tool that applies statistical and machine-learning algorithms

Is More Data Really Better?

Possibly, but beware of …

Data qualityPoor quality data will drive your analyses

Creating more noise, so that the signal is tougher to find

Feature extraction and feature selection are important steps that can really make a difference in the success of your algorithms

March 29, 2017 5

Page 6: Discovery Through Situational Awareness - NASPI€¦ · Discovery Through Situational Awareness. Project Goal. Create a tool that applies statistical and machine-learning algorithms

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.

March 29, 2017 [email protected] 6

Page 7: Discovery Through Situational Awareness - NASPI€¦ · Discovery Through Situational Awareness. Project Goal. Create a tool that applies statistical and machine-learning algorithms

Baselining Grid Behavior

Univariate 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 behavior

March 29, 2017 7

StaticBaselining Limits

Page 8: Discovery Through Situational Awareness - NASPI€¦ · Discovery Through Situational Awareness. Project Goal. Create a tool that applies statistical and machine-learning algorithms

Multivariate Baselining

Baseline captures what normal behavior is expected to be

Group similar behavior Time periods that group together indicate normal grid behaviorVariables that group together indicate highly correlated variables and may be candidates for feature reduction

Identify 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 variable

March 29, 2017 8

Page 9: Discovery Through Situational Awareness - NASPI€¦ · Discovery Through Situational Awareness. Project Goal. Create a tool that applies statistical and machine-learning algorithms

March 29, 2017 9

Creating a Baseline –Unsupervised Learning

Baselining Learning Algorithm

Model

Training Data:

Historical PMU Data

Real Time PMU Data

Class 1

Class 2

Class 3

Class 4

Class 5

Page 10: Discovery Through Situational Awareness - NASPI€¦ · Discovery Through Situational Awareness. Project Goal. Create a tool that applies statistical and machine-learning algorithms

March 29, 2017 10

Identifying Data Driven Atypical Events

Using multivariate statistical techniques to establish baselines of typical behavior, atypical moments in time can be discovered and the responsible variables can be identified.

Substation A

Substation B

Page 11: Discovery Through Situational Awareness - NASPI€¦ · Discovery Through Situational Awareness. Project Goal. Create a tool that applies statistical and machine-learning algorithms

Recent Atypicality Example

Processed 5 months of 60 frames/sec PMU data (15 PMUs)

This analysis only focused on phase angle pairs

Atypicality first detected at 6:23 and then atypicality occurred off and on for the next hour or so.

March 29, 2017 11

05:00 05:30 06:00 06:30 07:00 07:30 08:00

05

1015

20

Atypicality initially detected

All results have been de-identified

Atypicality Score

Page 12: Discovery Through Situational Awareness - NASPI€¦ · Discovery Through Situational Awareness. Project Goal. Create a tool that applies statistical and machine-learning algorithms

Atypicality Displays

Spatial plot showing which phase angle pairs were contributing to the atypicality (red indicates atypical)

March 29, 2017 12

Page 13: Discovery Through Situational Awareness - NASPI€¦ · Discovery Through Situational Awareness. Project Goal. Create a tool that applies statistical and machine-learning algorithms

Atypicality Displays

March 29, 2017 13

PMU1-PMU2 Pair PMU1-PMU6 PairPMU1-PMU3 Pair

2 of the 4 angle pairs that were significant.Angle pair was not significant, although it also contained PMU1.

Page 14: Discovery Through Situational Awareness - NASPI€¦ · Discovery Through Situational Awareness. Project Goal. Create a tool that applies statistical and machine-learning algorithms

Atypicality DetectionLightning Related Anomaly

March 29, 201714

Atypicality Score Substation A

Substation B

Page 15: Discovery Through Situational Awareness - NASPI€¦ · Discovery Through Situational Awareness. Project Goal. Create a tool that applies statistical and machine-learning algorithms

Atypicality DetectionAnomaly Related to Loss of Generation

March 29, 2017 [email protected] 15

Atypicality Score

Other PMUs behaved similarly

Zoomed in

Page 16: Discovery Through Situational Awareness - NASPI€¦ · Discovery Through Situational Awareness. Project Goal. Create a tool that applies statistical and machine-learning algorithms

Unsupervised LearningNeural Networks

• What are neural networks? • Machine learning models that can learn highly non-linear

behavior.

• Why we need neural networks? • For sufficiently large networks, performance becomes a

function of the amount of data.

Page 17: Discovery Through Situational Awareness - NASPI€¦ · Discovery Through Situational Awareness. Project Goal. Create a tool that applies statistical and machine-learning algorithms

Neural Networks for Power Grid Data

• Neural networks a perfect fit for power grid applications because:• We have access to a lot of data

(volume, high frequency).• Power grid behavior is highly non-

linear.

Example: Detecting Oscillations in Power System

• Oscillation events are rare and can be difficult to detect.

• Approach: Train a neural network (called autoencoder) to learn when the grid is stable.

Train

Detection

Signal Reconstruction

Anomaly Reconstruction

Page 18: Discovery Through Situational Awareness - NASPI€¦ · Discovery Through Situational Awareness. Project Goal. Create a tool that applies statistical and machine-learning algorithms

March 29, 2017 18

Supervised Learning

Baselining Learning Algorithm

Predictive Model

Training Data:

Historical PMU Data

Real Time PMU Data

Class 1

Class 2

Class 3

Class 4

Class 5

NormalVoltage Drop

SurgeMaintenance

Weather

Weather

Normal

Voltage Drop

SurgeMaintenance

Page 19: Discovery Through Situational Awareness - NASPI€¦ · Discovery Through Situational Awareness. Project Goal. Create a tool that applies statistical and machine-learning algorithms

Conclusions

The power grid community is in the infancy stages of applying statistical and machine learning algorithms.Care must be taken in determining which data should be used, how features can be extracted from the data, and selecting which features will provide insight.Initial results show that data driven anomalies can be identified using multivariate analyses techniques. Some of these anomalies correspond to actual events, but some do not.The potential is great in bringing valuable insight into the power grid community by applying multivariate statistics and machine learning techniques.

March 29, 2017 19