1 of 33

DASHBOARD FOR POWER OUTAGE FORECASTS FOR AN EMERGENCY RESPONSE TO HURRICANES

HENRY CUI

SHUANG GUO

AKSHAY SHETTY

Jul 12. 2023

Directed by Prof. CEFERINO and PRATEEK ARORA

2 of 33

Project Overview

Part 1

2

3 of 33

The goal of this project is to create a pipeline that uses weather forecast, land cover, population and nightlights data to predict power outages before a hurricane hits. The predictions will be presented through a dashboard to inform utility companies and communities about the risks of power outages.

Predicting Power Outages

Figure 1. Hurricane damages in Florida

“

3

4 of 33

POWER GRIDS

Figure 2. Power grid system showcase

4

5 of 33

HURRICANE IMPACTS

Figure 3. Power outage caused by Isaias

5

6 of 33

HURRICANE IMPACTS

​

Figure 4. Power outage in Florida

6

7 of 33

LITERATURE REVIEW

  • RECOGNIZED AS ONE OF THE PRIMARY DRIVERS OF LARGE-SCALE POWER OUTAGES…... (OUYANG ET AL., 2014)

​

  • THE HIGH SPEED WINDS AND HEAVY RAINFALL SHOWN WITH THE HURRICANE CAN DAMAGE POWER LINES, TRANSFORMERS, AND OTHER CRITICAL INFRASTRUCTURE…… (LIU ET AL., 2008) (HOU ET AL., 2022)

​

  • A MODEL DEVELOPED BY HOU ET AL. AIMED TO ANTICIPATE THE DURATION OF DISRUPTIONS AMIDST HURRICANES…... (HOU ET AL., 2023)

7

8 of 33

  • To understand the risk related to power grid for future scenarios based on our current understanding and available data to past events?

​

  • How can predicting modeling help?
    • Alternate sources of energy, when main grid is off
    • Rapid deployment of crew to repair the damaged sections of grid

Problem Definition

“

8

9 of 33

Utility Companies

Stakeholders

  • Installing backup power
  • Placing their crews ahead of a storm to recover the power system rapidly
  • Emergency Preparedness
  • Minimizing Disruption
  • Prioritizing Repairs

Community

“

9

10 of 33

Machine Learning

  • Mathematical models → future outcomes

​

Supervised Learning

  • Set of variables → dependent variable
  • The use of inputs to predict the outputs
  • regression analysis.
  • E.g. hurricane wind → power outages

​

Components of Machine Learning

  • Mathematical models
  • Measure of goodness

11 of 33

Inputs

Rainfall

Windspeed

Rainfall

Land cover

Historical Nightlights

Historical Power Outage

Output

Regression

Models

Probabilistic

Models

(Power Outage

Data for

Historical

Hurricanes)

“

11

12 of 33

Data Introduction

Part 2

12

13 of 33

GEOGRAPHY OF HURRICANES

  • Florida
  • Ian, Michael
  • Texas
  • Harvey
  • New Jersey
  • Isaias
  • Atlantic Basin

14 of 33

DATASETS

SOURCE:

NASA’S VISIBLE INFRARED IMAGING RADIOMETER SUITE (VIIRS) DAY/NIGHT BAND (DNB)

RESOLUTION:

500 M × 500 M

SOURCE:

NATIONAL CENTER FOR ATMOSPHERIC RESEARCH (NCAR)

RESOLUTION:

1 KM × 1 KM

SOURCE:

NATIONAL LAND COVER DATABASE (USGS NLCD)

RESOLUTION:

30 M × 30 M

SOURCE:

NATIONAL CENTER FOR ATMOSPHERIC RESEARCH (NCAR)

RESOLUTION:

1 KM × 1 KM

NIGHTLIGHTS

RAINFALL

WIND SPEED

LAND COVER

15 of 33

NIGHTLIGHTS

  • Provided by NASA
  • Available in A3(monthly/average) and A2(daily)
  • Originally in h5 format

​

​

A3 data of Panama City, FL

Figure 5. Nightlight in Panama City, FL

15

16 of 33

VALIDATION OF EFFECT FROM HURRICANES

Figure 6. Nightlight histogram of NJ, during Isaias attacked

17 of 33

WIND MODEL

  • Damage the poles and distribution lines
  • Tree damages
  • 3-sec gust wind speed
  • Strong winds → winds with speed over 20m/s

Figure 7. Wind distribution of NJ during Isaias attacked

18 of 33

  • Power grid management can vary
  • From USGS
  • Available in raster format
  • Resolution of 30m x 30m square pixels
  • 20 different classes
  • Input parameter is percent of area covered by different land classes in a city

LAND COVER

Figure 8. Florida land cover data

19 of 33

  • Damage to the utility infrastructure
  • Data from NASA GES DESC at 0.1 degree x 0.1 degree
  • Originally in NetCDF4 format

​

RAINFALL

Figure 9. Florida rainfall map during hurricane Ian attacked

20 of 33

POWER OUTAGE

  • Provided by pous
  • Recorded by counties
  • Available in csv format

​

Figure 10. NJ outages after Isaias attacked

20

21 of 33

Data Engineering and Analysis

Part 3

21

22 of 33

DATA ENGINEERING

VNP46A2/A3 (h5)

Illumination data (tiff)

Power indication (csv)

Nightlights

Rainfall

Land cover

Wind speed

NLCD database (tiff)

Land cover data (csv)

netCD4 data (cd4)

Precipitation data (csv)

3s gust wind speed

Wind data (csv)

Customer outage report

Available as csv file

Result

Preliminary process

In python

Zonal

statistic

Zonal + Clipping

Statistic 1111111111111

Process

In Python

Collected by

Unified grid system

Regression models

23 of 33

Missing values

Before

After

OUR APPROACH FOR MISSING DATA

Figure 11. Imputation results

24 of 33

LAND COVER DATA TRANSFORMATION

  • Calculated in 500x500 meters grid in QGIS

Figure 12. Land cover data processed

25 of 33

RAINFALL DATA IN TABLE

  • First two columns are calculated from and is coordinated with the grid
  • After that, the rainfall is specified in each square

26 of 33

Linear Regression

R2 = 0.149619

Random Forest

R2 = 0.554349

Gradient Boosting

R2 = 0.466477

XGBoost

R2 = 0.525253

LightGBM

R2 = 0.534762

Model Evaluation

REGRESSION RESULTS

Figure 13. Regression model results visualization

27 of 33

Best Performing Model

Figure 14. Regression results comparison

28 of 33

Summary

Part 4

28

29 of 33

LIMITATIONS

  • Prediction accuracy
  • Missing data
  • Data resolution

29

30 of 33

POLICY RECOMMENDATION

  • Resource Allocation
  • Public Alerts
  • Partnerships
  • Maintenance and update

30

31 of 33

NEXT STEPS

​

  • Connect with local department
  • Algorithm improvement
  • More case studies

31

32 of 33

REFERENCE

Ouyang, M., & Dueñas-Osorio, L. (2014). Multi-Dimensional Hurricane Resilience Assessment Of Electric Power Systems. Structural Safety, 48, 15-24. Doi:10.1016/J.Strusafe.2014.01.001

​

Liu, H., Davidson, R. A., & Apanasovich, T. V. (2008). Spatial Generalized Linear Mixed Models Of Electric Power Outages Due To Hurricanes And Ice Storms. Reliability Engineering & System Safety, 93(6), 897-912. Doi:10.1016/J.Ress.2007.03.038

​

Hou, H., Zhang, Z., Wei, R., Huang, Y., Liang, Y., & Li, X. (2022). Review Of Failure Risk And Outage Prediction In Power System Under Wind Hazards✰. Electric Power Systems Research, 210, 108098. Doi:10.1016/J.Epsr.2022.108098

​

Winkler, J., Dueñas-Osorio, L., Stein, R., & Subramanian, D. (2010). Performance Assessment Of Topologically Diverse Power Systems Subjected To Hurricane Events. Reliability Engineering & System Safety, 95(4), 323-336. Doi:10.1016/J.Ress.2009.11.002

​

Hou, H., Liu, C., Wei, R., He, H., Wang, L., & Li, W. (2023). Outage Duration Prediction Under Typhoon Disaster With Stacking Ensemble Learning. Reliability Engineering & System Safety, 237, 109398. Doi:10.1016/J.Ress.2023.109398

​

Arora, P., (2023). Probabilistic modeling of vulnerability of Power Grid to Hurricanes

​

33 of 33

THANKS!