Solar Panel Identification Under Limited Labels

Shuman Luo, Yang Weng, Elizabeth Cook, Robert Trask, Erik Blasch

Research output: Chapter in Book/Report/Conference proceedingConference contribution

1 Scopus citations


To better manage the unconventional two-way power flow, utilities are in urgent need to identify the locations of residential photovoltaic (PV) systems. With accurate PV location information, utilities can better maintain sustainable management and the safety of power grids. However, historical location records are unreliable and verification with human efforts is expensive, which challenges the PV location identification task. Thanks to the abundant information available from billing meter data, one can solve the problem via machine learning. As the labeled data can be very limited, supervised learning will be ineffective. Therefore, we propose new semi-supervised learning and one-class classification methods based on autoencoders. The proposed methods have been tested on a real-world utility data set and have shown superior detection accuracy in terms of accuracy and F1 score (= 0.95).

Original languageEnglish (US)
Title of host publication2022 IEEE Power and Energy Society General Meeting, PESGM 2022
PublisherIEEE Computer Society
ISBN (Electronic)9781665408233
StatePublished - 2022
Event2022 IEEE Power and Energy Society General Meeting, PESGM 2022 - Denver, United States
Duration: Jul 17 2022Jul 21 2022

Publication series

NameIEEE Power and Energy Society General Meeting
ISSN (Print)1944-9925
ISSN (Electronic)1944-9933


Conference2022 IEEE Power and Energy Society General Meeting, PESGM 2022
Country/TerritoryUnited States


  • autoencoder
  • detection
  • locations
  • one-class classification
  • semi-supervised learning
  • solar panels

ASJC Scopus subject areas

  • Energy Engineering and Power Technology
  • Nuclear Energy and Engineering
  • Renewable Energy, Sustainability and the Environment
  • Electrical and Electronic Engineering


Dive into the research topics of 'Solar Panel Identification Under Limited Labels'. Together they form a unique fingerprint.

Cite this