Zeal Shah

Ph.D. Student | Researcher, Data Science for Development


I am a fourth year Ph.D. student in the STIMA Lab (Systems Towards Infrastructure Measurement and Analytics), advised by Professor Jay Taneja. My research focuses on developing data-driven tools for measuring and mapping quality and reliability of electricity systems in a non-intrusive manner.

I develop both -- localized and wide-area -- monitoring solutions. Localized monitoring techniques involve use of cameras to measure phase and quality of electricity. The wide-area monitoring approaches combine large volumes of daytime and/or nighttime satellite images and a variety of ground-truth datasets (surveys, smart meter data, utility data), to estimate electrification status, degree of electrification, power supply inconsistencies and energy demand using satellite imagery.

My work mainly involves data analytics, geospatial data processing, machine learning, deep learning and signal processing.

Work Experience

Graduate Research Assistant

STIMA Lab, UMass Amherst | Aug 2018 - Present

  • Research and develop data-driven solutions to measure and map energy infrastructure systems in developed and developing settings. Please refer to the list of research projects and publications for more detail.

AI Engineering Intern (Remote)

Atlas AI | Aug 2018 - Present

  • Developed a Machine Learning (ML) model to detect electrification in satellite images and scaled it to produce high-resolution monthly electrification records for Africa from 2012-20 – a locational intelligence product offered by Atlas AI to assist its clients with strategically identifying sites for new infrastructure projects.
  • Built prototype of energy consumption prediction data layer that used electrification maps, publicly available satellite data and surveys to estimate tiers of energy consumption at a spatial resolution of 0.5km; aimed at helping clients locate clusters of target customers using granular insights into local energy demand.
  • Explored and ingested satellite data using Google Earth Engine, trained and evaluated the ML models using Python on Google Compute Engine, and stored the results in Google Bucket and BigQuery.

Data Science Intern

SparkMeter, Inc. | Feb 2018 - Sep 2018

  • Analyzed smart meter data to quantify the evolution of electricity quality, reliability and consumption across 68 sites in Africa with 200+ smart meters per site, for SparkMeter's keynote presentation at Microgrid Global Innovation Forum'18.
  • Created a Grafana dashboard using SQL scripting for real-time monitoring of deployed smart meters, base stations and cloud services to facilitate efficient and faster troubleshooting.
  • Provided need based data analysis and visualization support to different teams including engineering, hardware and customer success.

Engineering Intern (Remote)

Nikola Power | Jun 2018 - Jul 2018

  • Assisted in development of an optimal battery dispatching algorithm to minimize the operating cost of residential solar grid+storage system by controlling charging & discharging of the battery.

Graduate Teaching Assistant

Carnegie Mellon University | Jan 2017 - Dec 2017

  • Graduate TA for two senior level courses: Fundamentals of Power Systems and Embedded Systems.
  • Prepared and delivered technical lectures on modeling & simulation of power systems, and solving computational power systems problems like optimal power flow in MATLAB.
  • Helped 30 students with course material.

Data Science Intern

SparkMeter, Inc. | May 2017 - Aug 2017

  • Designed and developed smart meter data intelligence reports using Python to periodically provide utilities with actionable insights into commercial and technical operations of their grids.
  • Enhanced analytics and data visualizations delivered in SparkMeter’s data intelligence reports by incorporating feedback fromutility customers.

Research Projects

Monitoring Electric Grid Inconsistencies Using Satellite Data

  • Develop a novel, open-source, geo-spatial data-driven tool to enable high-resolution, spatio-temporal monitoring of electric grid reliability at global scale using nighttime-lights (NL) satellite data. A consistent, global grid reliability dataset doesn’t exist till date.
  • Demonstrated tool's application and limitations in detecting outages and estimating grid reliability at 0.5km spatial resolution in Accra, Ghana. Full paper submission under review.
  • Collaborators: UC Berkeley, Atlas AI, Colorado School of Mines.

Measuring Equity of Outages Using Satellite Data

  • Produce spatially-explicit historical records of outages using NL data to study the equity of outages across regions, and identifying disparities causing varied outage experiences
  • Published a case-study on the inequitable nature of outages observed during February 2021 snowstorm across Texas. Work done in collaboration with researchers from Lawrence Berkeley National Lab and Colorado School of Mines. Link

Deep Learning for Measuring Electrification Using Daytime Satellite Images

  • Supported on-going research efforts in energy access planning and energy demand forecasting by estimating electrification status and type (residential and non-residential) of structures in 9 million daytime satellite images spanning Kenya, using convolutional neural networks (CNN).
  • Accepted at ML4D workshop at NeurIPS 2021. In collaboration with Columbia University.
  • Extending the work to compare results with multiple open-source/private electrification datasets available in the market.

Mapping Disasters & Tracking Recovery in Conflict Zones Using Nighttime Lights

  • Developed a technique to generate high resolution (~0.5km) disaster maps of a region as soon as one day after the crisis begins, to deliver up-to-date information regarding spatial extent of the disaster to disaster managers.
  • Demonstrated a method to precisely identify the day when bombing occured by tracking sudden changes in nigttime lights illumination.
  • Proposed and evaluated two recovery measurement indices to trace recovery of the region spatially and temporally using nigttime lights data.
  • Published a full paper in the proceedings of IEEE GHTC'20. In collaboration with Colorado School of Mines.Link.

Monitoring Electricity Using Cameras and Visible Lights

  • Develop a non-intrusive grid sensing system using cameras to monitor power quality of the electric grids at the edges of the distribution network in real time; evaluate and improve the system to make it deployable at scale.
  • Demonstrated the use of cell phone and machine vision cameras to non-intrusively monitor electric-grid frequency and phase with errors of 1-2%, and 2-10%, respectively.
  • Developed a novel technique that uses cameras to passively monitor voltage, and obtained an error of 8-15% for measuring voltage of a light bulb that our system had not seen previously.
  • Published a full paper in the proceedings of ACM BuildSys'19. Link


Research Papers

  • Zeal Shah, Noah Klugman, Gabriel Cadamuro, Feng-Chi Hsu, Christopher D. Elvidge, Jay Taneja. “The Electricity Scene from Above: Exploring Power Grid Inconsistencies Using Satellite Data in Accra, Ghana.” Submission under review. (Full paper)
  • Zeal Shah, Simone Fobi, Gabriel Cadamuro, and Jay Taneja. “A Higher Purpose: Measuring Electricity Access Using High-Resolution Daytime Satellite Imagery.” Machine Learning for Development (ML4D) workshop at NeurIPS'21, December 2021. Ranked among top 3 papers*. (Workshop paper)
  • Santiago Correa, Zeal Shah, Jay Taneja. “This Little Light of Mine: Electricity Access Mapping Using Night-Time Light Data.” In the proceedings of the 12th International Conference on Future Energy Systems (ACM e-Energy'21), June 2021. Link. (Short paper)
  • Zeal Shah, Feng-Chi Hsu, Christopher D. Elvidge, and Jay Taneja. “Mapping Disasters & Tracking Recovery in Conflict Zones Using Nighttime Lights.” In the IEEE Global Humanitarian Technology Conference (GHTC'20), October 2020. Link (Full paper)
  • Zeal Shah, Alex Yen, Ajey Pandey, and Jay Taneja. “GridInSight: Monitoring Electricity Using Visible Lights.” In the 6th ACM International Conference on Systems for Energy-Efficient Built Environments, Cities, and Transportation (BuildSys’19), November 2019. Best Paper Nominee. Link (Full paper)

Book Chapters

  • Zeal Shah, Ali Moghassemi, and Panayiotis Moutis. “Frameworks for Considering RES and Load Uncertainties in VPP Decision-Making.” Scheduling and Operation of Virtual Power Plants by Elsevier. To be published in 2022.

Blogposts / Notes

  • Zeal Shah, JP Carvallo, Feng-Chi Hsu, Jay Taneja. “Frozen Out in Texas: Blackouts and Inequity”. End Energy Poverty Field Note published by the Rockefeller Foundation, April 2021. Link
  • Stephen Lee, Zeal Shah, Brian Min, Jay Taneja. “Lighting the Way: Nighttime Lights for Electrification Planning”. Memo published by Energy for Growth Hub, April 2021. Link
  • Zeal Shah. “What is Temporal Resolution?". Blogpost published by Atlas AI, June 2020. Link

Posters & Presentations

  • Zeal Shah, Jay Taneja. “Exploring Power Grid Inconsistencies Using Satellite Data in Accra, Ghana.” UMass Energy Forum, April 2021. Link
  • Zeal Shah, Jay Taneja. “Monitoring Electric Grid Reliability Using Satellite Data.” In the 6th ACM International Conference on Systems for Energy-Efficient Built Environments, Cities, and Transportation, November 2019. Best Poster Award. Link
  • Zeal Shah, Alex Yen, Ajey Pandey, Jay Taneja. “GridInSight: Monitoring Electricity Using Visible Lights.” In the 2nd Annual ACM SIGCAS Conference on Computing and Sustainable Societies (COMPASS’19), July 2019. Link
  • “Smart Metering Data For Tracking Access to Electricity.” SparkMeter's keynote at the 7th Microgrid Global Innovation Forum, September 2018. (Presented by Jon Thacker)
  • Zeal Shah, Yoolhee Kim, Anand Prakash, Vasu Nambeesan. “Occupancy Prediction Based on the Power Consumption Patterns.” In the Carnegie Mellon University Symposium on Machine Learning in Science and Engineering, May 2017. Link
  • Zeal Shah, Siddhartha Joshi. “Operation and Analysis of a Bi-directional DC-DC Converter for Efficient Charge Control of Battery in a Microgrid.” In the 50th IEEE Industry Applications Society Annual Meeting, October 2015.

Selected Course Projects

Multi-tier Online Book Store

Developed a multi-tier web application using Flask in Python and added features like caching, replication, load-balancing, fault tolerance and recovery. Link

Where, When and Watt?

Created a program to predict occupancy of different rooms based on appliance power consumption data and achieved 93% model prediction accuracy. Link

New York State Energy Brief

Analyzed multiple open-source datasets to study and predict NY’s energy consumption in residential, commercial, industrial and transportation sectors.

Solving Unit Commitment

Implemented mixed-integer linear programming (Branch & Bound method) to solve a 24-hour unit commitment problem using data from multiple generators and demand data.

Energy Policy & Economics

Studied and delivered policy evaluation write-ups on energy security, clean power plan, emissions from internationaltransportation, 100% renewable vs. 0 emissions, and renewable integration in regulated vs. deregulated electricity markets.

Analysis of Energy Transport & Storage in Switzerland

Studied and presented a detailed analysis of energy transmission and storage in Zürich and Canton of Valais; Proposed solutions to reduce congestion and achieve their future energy goals