Projects at UMass Amherst

This Image is from NOAA webpage

Solar Forecasting using Multispectral GOES-R Satellite Data

GOES-R range of satellite data show promising results for solar modeling and forecasting (as shown in our prior work). It will be very interesting to use it for predicting solar generation at near-future time instants. We are working on developing machine learning models for the same.

This is presently a work in progress.

Solar Modeling via Multispectral Channel Data from GOES-R Satellites

Satellite data is important for variety of socially relevant applications, one of which being solar modeling. With the new range of GOES -r series of satellite available, there is whole new domain of information which can be used for various purposes. We have used data from GOES-16 and GOES-17 satellites in the following projects -

  • NOAA releases information in 16 different spectral bands at frequency of 5 minutes in the form of channel data ranging at a range of 0.5-2km. Each of these channel addresses a wavelength range and subsequently can be used for different purposes. We use information from the channels of both GOES-16 and GOES-17 to model the solar output directly using machine learning algorithms. Published in BuildSys 2020.

Data and the codes will be released soon.

This Image is from NOAA webpage

Solar Modeling via Ground Level and Secondary Derived Products (DSR) by GOES-R Satellite

We show, the accuracy of solar modeling using satellite data (DSR), public weather station data i.e. cloud conditions (Okta) and SURFRAD (Surface Radiation Budget) Network. We also analyze some hybrid approaches using machine learning to combine the best of both ground and satellite approaches. We show our results on a set of ~50 solar sites in the United States. To the best of our knowledge this is the first time DSR data from NOAA has been used for solar modeling.

Published in SmartGridComm 2020.

Data and the codes will be released soon.

We explore using a mixture of solar, batteries, and a whole-home natural gas generator to shift users partially or entirely off the electric grid. We assess the feasibility and compare the cost and carbon emissions of such an approach with using grid power, as well as existing “net metered” solar installations. Our results based on analysis done on many solar sites across U.S. show that the approach is trending towards cost-competitive based on current electricity prices, reduces carbon emissions relative to using grid power, and enables users to install solar without restriction.
Published in SmartGridComm 2019

We proposed Software-defined Solar- powered (SDS) systems that dynamically regulate the amount of solar power that flows into the grid. To enable SDS systems, our work introduced fundamental mechanisms for programmatically controlling the size of solar flows, including mechanisms to both enforce an absolute limit on solar output and a new class of Weighted Power Point Tracking (WPPT) algorithms that enforce a relative limit on solar output as a fraction of its maximum power point (MPP). We implemented an SDS prototype, called SunShade, and evaluated tradeoffs in the accuracy and fidelity of these mechanisms to enforce limits on solar flows.
Published in ICCPS 2017.

Previous Projects

NTU Singapore - as Research Associate

Cell Balancing Strategy for Battery Management

Design and implementation of cell balancing strategy algorithm for smart batteries.

Mixed-criticality RTOS based on LITMUSRT

Worked on developing and testing the scheduling algorithms on mixed criticality based applications for mutli-core processors.

IIIT Bangalore - as part of MS thesis

Distributed Health Monitoring system for Smart Grid

  • Developed programmable system for distribution level of grid failure prevention.

  • Proposed and tested grid classification system for different stability zones.

  • Proposed communication architecture for smart meter and tested it to send last gasp messages.
    Multiple publications listed under Publications page.