Energy Data Analytics Lab (2015-2016)
There are increasingly many sources of energy data in the modern world. Examples include electric utility smart meters capable of providing minute-by-minute data from millions of buildings, power market data tracking real-time energy prices across the nation, satellite imagery of energy systems as well as thermostat and control system sensor data in buildings, among others. Based at the Duke University Energy Initiative, the Energy Data Analytics Lab’s mission is to develop and apply data analytics tools to transform diverse energy data into innovative solutions for increasing the reliability, security, resilience and environmental sustainability of energy systems, while reducing costs.
Through team-based problem-solving, members of this Bass Connections project team learned to describe the potential effects of data analytics tools on energy systems, mastered the fundamentals of machine learning and big data algorithms and began to implement and understand the implications of results from experimental research techniques.
Increased attention to the global warming crisis has led to the rapid adoption of distributed rooftop solar photovoltaics across the U.S. Simultaneously, object detection in imagery is now a common research tool with varying applications ranging from military surveillance to facial recognition on social media. With the recent proliferation of solar panels, remote object detection has become an increasingly attractive tool to help track solar installations nationwide.
Using a ground-truth dataset of over 19,000 solar arrays in Fresno, Modesto, Oxnard and Stockton, California, amassed by the 2014-2015 version of this project team, the 2015-2016 team developed a solar photovoltaics identification algorithm with 90.4% accuracy when extracting regions using the maximally stable extremal regions (MSER) technique, and with 99.5% accuracy when using regions formed from the area surrounding solar arrays in the ground-truth training set.
Because both methods share preprocessing, feature extraction and classification techniques, but differ in region extraction, the team identified region extraction to be the key cause of the discrepancy, and thus, the main constraint in the overall PV identification process.
In Summer 2016, a Data+ team took this work further by building a ground truth dataset comprising satellite images, building footprints, and building heights (LIDAR) of more than 40,000 buildings, along with road annotations. This dataset can be used to train computer vision algorithms to determine a building’s volume from an image, and is a significant contribution to the broader research community with applications in urban planning, civil emergency mitigation and human population estimation.
Fall 2015 – Spring 2016
Jordan M. Malof, Kyle Bradbury, Leslie M. Collins, Richard G. Newell, Alexander Serrano, Hetian Wu, Sam Keene. “Image Features for Pixel-Wise Detection of Solar Photovoltaic Arrays in Aerial Imagery Using a Random Forest Classifier.” 2016. Institute of Electrical and Electronics Engineers International Conference on Renewable Energy Research and Applications:799-803.
Jordan M. Malof, Leslie M. Collins, Kyle Bradbury, Richard G. Newell. “A Deep Convolutional Neural Network and a Random Forest Classifier for Solar Photovoltaic Array Detection in Aerial Imagery.” 2016. Institute of Electrical and Electronics Engineers International Conference on Renewable Energy Research and Applications:650-654.
Jordan M. Malof, Kyle Bradbury, Leslie M. Collins, Richard G. Newell. “Automatic Detection of Solar Photovoltaic Arrays in High Resolution and Aerial Imagery.” 2016. Applied Energy 183:229-240
Kyle Bradbury, Raghav Saboo, Timothy L. Johnson, Jordan Malof, Arjun Devarajan, Wuming Zhang, Leslie M. Collins, Richard G. Newell. 2016. “Distributed Solar Photovoltaic Array Location and Extent Dataset for Remote Sensing Object Identification.” Scientific Data 3.
Automated Rooftop Solar PV Detection and Power Estimation through Remote Sensing (poster by Arjun Devarajan, Brody Kellish, Cassidee Kido, Aaron Newman, Nick Von Turkovich, Wuming Zhang, Kyle Bradbury, Leslie Collins, Timothy Johnson, Richard Newell)
Bradbury, K.; Brigman, B.; Collins, L.; Johnson, T.; Lin, S.; Newell, R.; Park, S.; Suresh, S.; Wiesner, H.; Xi, Y. 2016. Aerial Imagery Object Identification Dataset for Building and Road Detection, and Building Height Estimation.
Bradbury, K.; Saboo, R.; Johnson, T.; Malof, J.; Devarajan, A.; Zhang, W.; Collins, L.; Newell, R. 2016. Distributed Solar Photovoltaic Array Location and Extent Data Set for Remote Sensing Object Identification. (data set of known solar panel locations in imagery data containing over 18,000 hand-annotated locations of solar panels and their boundaries)
This Team in the News
- Kyle Bradbury, Pratt School of Engineering-Electrical & Computer Engineering|Energy Initiative
- Leslie Collins, Pratt School of Engineering-Electrical & Computer Engineering
- Timothy Johnson, Nicholas School of the Environment-Earth and Climate Sciences
- Richard Newell, Nicholas School of the Environment-Environmental Sciences and Policy
- Steven Palumbo, Duke Facilities Management
/graduate Team Members
Yixuan Zhang, Master of Environmental Management, Energy and Environment
/undergraduate Team Members
Chia Rui Chang, Statistical Science (BS)
Arjun Devarajan, Computer Science (BS)
Brody Kellish, Electrical & Computer Egr(BSE), Computer Science (BSE2)
Cassidee Kido, Electrical & Computer Egr(BSE)
Ting Lu, Economics (BS)
Aaron Newman, Electrical & Computer Egr(BSE), Computer Science (BSE2)
Nicholas Von Turkovich, Electrical & Computer Egr(BSE), Computer Science (BSE2)
Wuming Zhang, Computer Science (BS), Statistical Science (BS2)
/yfaculty/staff Team Members
Guillermo Sapiro, Pratt School of Engineering-Electrical & Computer Engineering