Energy Data Analytics Ph.D. Student Fellows Program

Simachew Ashebir is a Ph.D. student in data science and analytics at North Carolina A&T State University’s College of Science and Technology. His research interest is time-series data forecasting.

Elizabeth Brown is a Ph.D. student in public policy at the University of North Carolina at Chapel Hill’s College of Arts and Sciences. Her research focuses on designing policy interventions to reduce energy poverty with low-carbon and renewable technologies.

Katherine Burley Farr is a Ph.D. student in public policy at the University of North Carolina at Chapel Hill’s College of Arts and Sciences. Her research interests include energy and climate policy, in particular, subnational policy responses to climate change and the renewable energy transition.

Amanda Gregg is a Ph.D. student in materials science and engineering at Duke University’s Pratt School of Engineering. She is working on developing energy-efficient artificial lighting technologies using metamaterials.

Zehao Jin is a Ph.D. candidate in environmental engineering at Duke University’s Pratt School of Engineering. His major research interest is using machine learning and spatial data analysis techniques to develop an estimation system for the reuse potential of coal combustion byproducts.

Cameron Lisy is a Ph.D. student in operations research at North Carolina State University’s Colleges of Engineering and Science. His research objective is to provide decision support tools for the complete decarbonization of the US energy system and to help policy makers better understand the hidden risks to the electric power grid not considered with standard tools and practices.

Ying Yu is a Ph.D. student in environmental science and engineering at the University of North Carolina at Chapel Hill’s Gillings School of Public Health. She is interested in using microeconomic methods to answer research questions related to climate and energy, with a particular focus on equity discussions.

Andrew is a Ph.D. student in economics at North Carolina State University. His research focuses on analyzing the effects of renewable energy and renewable energy technologies on the food-energy-water nexus, identifying efficient electricity market mechanisms in the face of climate change and the continued rise of renewables, and characterizing the interplay between the effects of both electric vehicles and electric vehicle charging on the industrial organization of the electricity sector.

Project: Parched Power Plants: The Role of Markets and Plant Traits in Power Plants’ Drought Response
Climate change intensifies a myriad of climatic events (such as drought) that then affect key economic inputs (e.g., water) and outputs (e.g., electricity). Drought diminishes the water supply that thermoelectric power plants rely on for cooling purposes, thereby forcing plants to adjust their operations along several margins and stressing the supply of electricity. Studies of large-scale, aggregated, or short-term electricity sector responses have provided valuable insights into the immediate plant-level relationships between drought and electricity production, but not much is known about the implications of electricity market structure or water rights systems for plant-level drought and climate change resiliency. This project aims to leverage detailed panel data on plants’ operations and several proxies for drought (e.g., the Palmer Drought Severity Index, satellite data on groundwater concentration) in order to empirically characterize plants’ drought responses using econometric models and machine learning. A central goal and contribution of the project is to determine whether there are any plant efficiency or stability advantages to being in an electricity market area or possessing certain water rights. We also aim to predict plants’ responses under alternative policy and market scenarios, identify what factors drive plants’ responses, and discern any intra-or inter-plant substitution across generation methods and/or cooling systems.

Advisors: Dr. Eric Edwards and Dr. Jordan Kern.

Kshitiz is a Ph.D. candidate at the Department of City and Regional Planning at the University of North Carolina at Chapel Hill. His major research interest is using novel machine learning and geospatial data science techniques to help solve global challenges in planning decentralization and decarbonization of energy systems.

Project: Improving energy infrastructure detection from satellite imagery using Conditional Generative Adversarial Networks
Up-to-date information on energy infrastructure at the global level is inadequate. Although novel machine learning applications such as object detection on satellite imagery offer potential to detect such infrastructure, there is a lack of studies and training datasets at various geographies, especially in developing countries. Developing object detection models that generalize better to different geographies and resolutions of imagery would address the gap. To that end, we propose the development of Conditional Generative Adversarial Networks (CGAN)specifically designed to detect energy infrastructure (here, solar panels and pylons) and compare the CGANs to state-of-the-art object detection models (specifically Faster R-CNN). CGANs are chosen based on the promise shown by few CGAN based models in improving generalization across satellite imagery of different geographies. The study will help explore the potential of CGANs in improving energy infrastructure detection across geographies in order to create up-to-date information base and help energy infrastructure planning and achievement of development goals using easily available satellite imagery.

Advisors: Dr. Nikhil Kaza and Dr. Noah Kittner.

Bander is a Ph.D. student in mechanical engineering and materials science at Duke University. He is interested in developing robust data inverting models of atomic correlations impacting the performance of energy materials.

Project: Machine learning to accelerate energy materials research from neutron scattering experiments at Oak Ridge National Laboratory
Impacting the performance of energy materials, ranging from photovoltaics to solid-state batteries, are slow correlations of the motion of the atoms constituting them. National facilities such as Oak Ridge National Lab lead the path in probing the atomic correlations and extracting physical insights of promising candidate materials to advance the energy infrastructure. Therefore, it becomes a natural extension to utilize machine learnable models trained to invert the large, gathered datasets to accelerate the analysis of the atomic correlations impacting energy materials.

Advisors: Dr. Kyle Bradbury, Dr. Jordan Malof, and Dr. Olivier Delaire.

Rixi is a Ph.D. student in electrical and computer engineering at Duke University. He is interested in using data-driven methods to accelerate the solving of the electromagnetic inverse scattering problem.

Project: Data-driven Modeling and Design of Energy-efficient Metasurface Holograms
As the market of virtual reality/augmented reality (VR/AR) keeps exponentially growing and the concept of metauniverse emerges, computer-generated holography is becoming the enabling technology for creating a realistic three-dimensional(3D) lightfield environment. However, current hologram implementations are usually bulky in size and require high-power light source, which hinders wider adoptions of this technique. Besides, solving the forward and inverse design problem of the hologram is computationally intensive, which also prevents mobile holographic display with limited computation power. This project proposes to fully utilize the indispensable power of deep neural networks and all-dielectric metasurface to circumvent the limitations of computer-generated holography and to eventually pushforward VR/AR display to the next level. All-dielectric metasurface is artificial electromagnetic device which has been demonstrated to implement 3D holograms with high efficiency. The complexity of designing the energy-efficient metasurface hologram can be drastically reduced by the introduction of deep neural networks, which has already found successful applications in the electromagnetic design problems. Using deep neural networks for metasurface hologram design, in this project, is expected to bring down the obstacles of both compactness and high-power consumption of holograms, which paves the way for more widespread usages of holographic displays in a VR/AR future.

Advisors: Dr. Willie Padilla and Dr. Jordan Malof.

Jethro is a Ph.D. student in civil engineering at North Carolina State University focusing on energy systems modeling by use of the capacity credit metric. His work addresses innovative infrastructure development to improve societal quality of life and efficiency in the workplace.

Project: Resource Adequacy in Energy System Models
Revamping the electric grid to get to net-zero greenhouse gas emissions by 2050 has become an important international goal to avoid life-threatening environmental damage. Renewable energies are a proven, effective substitute for traditional carbon-dense fuels, but it will require effort and ingenuity to incorporate renewables at a scale that can effectively reduce emissions. The presence of weather risks especially during times of extreme natural events will yield correlated failures and affect grid reliability. By use of a metric called capacity credit, this project will help energy system planners better understand the benefit of renewables to the multi-regional grid of the U.S. while accounting for existing traditional generation.

Advisors: Dr. Jeremiah Johnson, Dr. Anderson Rodrigo de Queiroz, and Dr. Jordan Kern.

Alexander is a Ph.D. Candidate at North Carolina State University’s Center for Geospatial Analytics. With expertise in systems modeling and geospatial analytics, he researches questions at the intersection of energy, transportation, and land-use change.

Project: A Tool for Assessing Electric Vehicle Infrastructure Capacity across Emergency Evacuation Scenarios
Electric vehicles can play an important role in reducing diffuse greenhouse gas emissions and increasing the energy security of the United States. However, as the frequency and intensity of extreme weather events increases over time with climate change, so too does the need to think strategically about the future of fueling infrastructure for plug-in electric vehicles. Electric vehicles take much longer to refuel than conventional fuel vehicles (like gasoline and diesel), and so they represent a unique challenge when thinking about resilience and evacuation planning. Simultaneously, growth in the market for electric vehicles threatens to compound these issues as electric vehicles become a larger proportion of all vehicles on the road. To tackle this challenge, I am creating a traffic flow model that will help regional and transportation planners as well as electric utilities think about and plan around future electric vehicle adoption. By simulating different possibilities of electric vehicle infrastructure under varying evacuation scenarios, we can identify best practices and opportunities for improvement when it comes to the resilient design and siting of electric vehicle charging equipment. This knowledge can help both to keep drivers safe in emergency scenarios as well as encourage confidence and investment in electric vehicle charging infrastructure.

Advisors: Dr. Jelena Vukomanovic and Dr. Lindsey Smart.

Yang is a Ph.D. student in electrical and computer engineering at Duke University. He is interested in accelerating inverse design of revolutionary energy materials based on all-dielectric metasurfaces through deep learning. 

Project: Deep Learning for the Inverse Design of Metasurface-based Energy Materials 
All-dielectric metasurfaces emerged as prominent platforms to manipulate energy at surfaces and have proven their capabilities as highly efficient energy materials. With thermal and waste energy eliminated by tailored emission, all-dielectric metasurfaces provide a revolutionary solution for efficient energy harvesting. However, all-dielectric metasurfaces require extensive numerical simulations to understand their physical properties. Deep learning approaches have been studied on the accelerated inverse design of metasurfaces, yet more complex geometries remain mostly unexplored. I am interested in identifying deep learning solutions that can accurately identify complex all-dielectric metasurface geometry that yields targeted frequency-dependent scatterings. The project’s success should accelerate the inverse design of all-dielectric metasurfaces, offering unprecedented high-performance thermal materials in energy harvesting applications. The deep learning methods can particularly identify optimal ADM thermal emitters ready to deploy in the realization of high-efficiency thermophotovoltaic cells. 

Advisors: Dr. Jordan Malof, Dr. Willie Padilla.

Qian is a Ph.D. candidate in environmental engineering at North Carolina State University. Her research focuses on developing strategies to mitigate negative human health impacts associated with power sector emissions by studying the interactions between power sector emissions, air quality, and human health.  

Project: Reducing health impacts from the power sector emissions in China with economic dispatch and energy storage  
China has started a new round of power market reform, introducing a dispatch approach that minimize the electricity generation costs to its current power sector. Although several studies have investigated the carbon emission impacts of adopting this economic dispatch in China, none have estimated the human health impacts brought by this transition. Comprehensively understanding the impacts of the power market reform will provide insights on how to make better regulations to protect the public health. This project will estimate the health impacts by integrating power system models and air quality models, and also explore how to cost-effectively reduce these health impacts by internalizing real-time health costs in plant dispatch decisions and re-optimizing the unit commitment and economic dispatch in light of these impacts.  

Advisors: Dr. John Baugh, Dr. Fernando Garcia Menendez, Dr. Jeremiah Johnson.

Suhas Raju is a Ph.D. student in electrical engineering at the University of North Carolina Charlotte. His research focuses on energy efficiency through the use of artificial intelligence techniques, seeking to identify solutions to reduce greenhouse gas emissions in the commercial building sector. 

Project: Exploration of Neural Networks and Kalman filters to estimate parameters driving energy use.  
The buildings sector accounts for approximately 33% of worldwide greenhouse gas emissions and more than 40% of primary energy usage. Over the past 3 years we have analyzed over 6,000 small commercial buildings and identified issues driving high energy usage. Achieving only a 1% reduction in each of those 6,000 facilities would represent the removal of about 380 residential homes from the grid. The work in this project will focus on developing appropriate artificial intelligence approaches to leverage this large data set to detect reasons for high energy consumption. 

Advisors: Dr. Ahmed Arafa, Dr. Robert Cox.

Josh is a Ph.D. candidate in parks, recreation, and tourism management at North Carolina State University. He is a geographer interested in using spatial analysis to realize the equitable access of resources in society, particularly through informing policy and community action. 

Project: Scales of Energy Poverty: Understanding Energy Justice Through Spatial Analysis 
Nearly 1.4 million North Carolinians and 4 billion people around the world experience energy poverty - they are unable to both secure energy and acquire other basic human needs. Many community agencies, policy programs, and research studies identify energy poverty through the use of geospatial technologies and data. However, these efforts can fail to recognize many who need assistance due to the aggregation of spatial data to inappropriate scales. Because energy poverty is felt at a household level, aggregation to a large spatial scale (a county, for example) effectively hides forms of energy poverty, leading to many not receiving the type of help they need. My work will provide a suite of data analyses that identify spatial patterns of energy poverty hidden when data is aggregated answering who is missed and where. My goal is to use these outcomes to inform policymakers on the equitable outcomes of potential interventions, while also community leaders with information to combat energy poverty. 

Advisors: Dr. Christopher Galik, Dr. Jelena Vukomanovic.

Ben is a Ph.D. student in electrical and computer engineering at Duke University. Working in Duke’s Applied Machine Learning Lab, he is interested in applying advanced deep learning techniques to develop novel algorithms that can extract energy systems information from unmanned aerial vehicle (UAV) imagery. 

Project: Mapping energy infrastructure with UAVs and deep learning 
As energy systems undergo a dramatic transition to more renewable and distributed energy generation, energy security in the forthcoming decades will depend heavily upon increasingly sophisticated energy systems modeling and effective decision-making.  Success in these endeavors depends crucially upon access to high-quality and detailed information about existing energy infrastructure: e.g., residential solar systems and their capacity, transmission/distribution lines, and more. Unfortunately, however, such information is often limited, incomplete, or only accessible for a substantial fee. In this work, I propose to overcome this obstacle to energy decision-making by leveraging recent breakthroughs in machine learning to develop algorithms that can automatically extract energy information from high-resolution imagery from unmanned aerial vehicles (UAVs).  In recent years such algorithms have been successfully demonstrated for collecting energy information on satellite imagery, however, many types of energy infrastructure cannot be reliably identified or characterized using satellite imagery due to its limited resolution.  In this work, I propose to collect UAV imagery of several types of energy infrastructure and demonstrate that these objects can be identified and characterized with unprecedented accuracy.  This technique, if effective, will provide a powerful new tool that will facilitate energy modeling and decision-making, and help ensure energy security in the forthcoming decades.     

Advisors: Dr. Kyle Bradbury, Dr. Jordan Malof.

Celine is a Ph.D. candidate in civil and environmental engineering at Duke University. She is interested in applying deep learning techniques and Bayesian statistics to model and assess natech risk in complex and interconnected systems.   

Project: Detection of Above Ground Storage Tanks Using Faster R-CNN 
Energy systems are essential for society's functioning and economy; however, they are susceptible to natural hazard-induced failures. Extreme hydrological events can cause above ground storage tank (ASTs) floatation, bucking, and sliding failures, resulting in chemical releases that can migrate offsite posing threats to surrounding communities. Unfortunately, infrastructure fragility assessments, which required AST location, type, and volume, have been constrained by limited, incomplete, or inaccessible data. The increasing abundance of high-resolution overhead imagery collected through the National Agriculture Imagery Program (NAIP) may provide a valuable information source.  I propose to utilize deep learning methods to create multi-class object detection models and new above ground storage tank datasets for use in natural hazard risk assessment. 

Advisors: Dr. Mark Borsuk, Dr. David Carlson.

Zhenxuan is a Ph.D. student in the University Program of Environmental Policy at Duke University with an economics concentration. Lying at the intersection of environmental economics, climate change, and industrial organization, his research employs innovative data and methods to explore human and firm behavioral responses to environmental changes, and evaluate the efficiency and distributional effect of environmental policies. 

Project: Using Machine Learning Approach to Improve the Understanding of Household Adoption of Energy-Efficient Technologies 
The rapid growth in world energy consumption brings increased local air pollution and greenhouse gas emissions. Energy-efficient technologies can create substantial economic and environmental benefits by saving the financial costs and mitigating the environmental damages induced by rising energy consumption. Household adoption rate of these cost-effective technologies, however, is still low especially in developing countries due to several market failures and barriers. The increasing availability of big data on consumer choices provides both opportunities and challenges for studying household preferences for energy-efficient technologies. Using a comprehensive air conditioner transaction dataset, my project will develop an empirical method by combining machine learning approach with economic modeling of consumer demand to better understand household adoption of energy-efficient products and improve the out-of-sample prediction of their choices. 

Advisors: Dr. Billy Pizer, Dr. Cynthia Rudin.

Alina is a computer science Ph.D. student working with Professor Cynthia Rudin in the Prediction Analysis Lab. She is interested in identifying buildings with poor insulation to better inform civic planning as well as data concerning efficiency.  

Project summary: “A poorly insulated building requires more energy to heat and cool than a well-insulated building. Using publicly available thermal satellite imagery and a super-resolution algorithm, estimates can be made about the energy efficiency of individual buildings. Super-resolution is the problem of finding a high-resolution image from a low-resolution image. The goal is to create high-resolution, accurate, realistic reconstructed images from low-resolution images to enhance image data and recover visual information. By identifying buildings with poor insulation (and therefore poor energy efficiency), policy can be informed by data, such as providing incentives to install new insulation. My goal is for data to detect in-need communities not only by region, but by specific household, to further direct policy and civic planning.”

Bohao Huang is a Ph.D. student in electrical and computer engineering at the Pratt School of Engineering.  Working in Duke’s Applied Machine Learning Lab, he is interested in leveraging advances in deep learning to develop algorithms that can automatically extract energy systems information from aerial imagery.

Project Summary: “Energy security is vital to the prosperity and sustainability of modern societies. Ensuring energy security relies upon effective decision-making and energy systems modeling, a crucial component of which is access to high quality energy systems information. Unfortunately, however, such information is often of limited availability, incomplete, or difficult to access because it is proprietary. Aerial imagery (e.g., color satellite imagery) is increasingly cheap and abundant, and may provide a rich source of energy systems information, however extracting useful information from such imagery is costly. I propose to leverage recent advances in deep learning to develop algorithms that can automatically extract useful energy systems information from large volumes of aerial imagery, potentially yielding a powerful and scalable new source of such information.”

Jun is a Ph.D. student in earth and ocean sciences at the Nicholas School of the Environment. Her research models energy systems in the context of trade to better understand international energy security. 

Project Summary: “Embodied energy is the sum of energy consumed to produce a commodity and is widely used to assess the total environmental impacts of a production process. My project uses a hybrid input-output approach to model embodied energy flows throughout the global economy. This approach deviates from the existing literature by addressing the feedbacks between energy and non-energy industries rather than simply the use of energy by non-energy industries. This tool will allow for examination of industry requirements for energy transformations, as well as the embodied energy in supply chains, of over 130 countries across 20 years (1995-2015). My goal is to then use machine learning techniques to simulate the model for future impacts.”

Tongshu Zheng is a Ph.D. student in civil and environmental engineering at Duke’s Pratt School of Engineering. He is interested in leveraging data-driven techniques to solve air pollution problems including developing a novel satellite-based remote sensing algorithm to cost-effectively and accurately assess the loss in solar energy production due to particulate matter air pollution.

Project Summary: “Atmospheric particulate matter (PM) can diminish solar energy production through ambient PM and PM deposited on solar panels. Solar energy loss is estimated to be as high as 25–35% in highly polluted areas. My project aims to assess and predict the loss in solar energy production in a low-cost and accurate manner by using image processing techniques and machine learning algorithms to automatically extract information from the global, 3m resolution satellite imagery. I hope to inform solar energy companies about their daily production loss due to PM pollution as well as help them strategize the optimal solar panel cleaning frequency. I want my research to catalyze the development of startup companies that specialize in providing solar production loss data.”

Qingran Li is a Ph.D. student in the University Program in Environmental Policy (economics track) offered jointly by Duke's Nicholas School of the Environment and Sanford School of Public Policy. Her research includes using analytical tools to understand behavioral responses to policies and developing interdisciplinary solutions to energy and environmental issues.

Project summary: "Residential electricity consumption is an important indicator of household characteristics, but it is often held confidential by utilities and seldom reported by publicly available energy surveys. Missing such information significantly constrains our ability to answer important policy questions. My project targets a big question: How can we estimate residential electricity demand more precisely? Using the smart meter data set from an Irish CER trial project and the national time use survey, this project aims at correcting the estimation bias from behavioral and policy-related factors which are often overlooked in the conventional engineering and statistical models. A new algorithm will be developed to identify residential usage patterns with additional information provided by behavioral surveys so that information lost from inadequate load samplings can be compensated."

Edgar Virguez is a student in the doctoral program in environment at Duke's Nicholas School of the Environment. He is interested in contributing to the understanding of market mechanisms that facilitate the integration of variable energy resources. He holds a BS in environmental engineering, a BS in chemical engineering and a MS in environmental engineering. During the last decade, he has worked with several institutions (e.g., Universidad de los Andes, World Bank, Inter-American Development Bank) promoting the adoption of cleaner fuels in transport and industry throughout Latin America.

Project summary: "My project aims to design quantitative tools supporting the process of assessing policy and market approaches, promoting an increased penetration of variable energy resources in the energy matrix. This assessment will be performed based on the economic, reliability, and environmental dimensions of the electric power system, accounting for the benefits of reduced fuel use and emissions, and for the increased capital costs of renewables and the necessary re-dispatching of conventional generators."

Tianyu is a Ph.D. student in the department of computer science at Duke. Before coming to Duke, he obtained a BS in mathematics and computer science from the Hong Kong University of Science and Technology. His general research interests are in machine learning and applications of machine learning algorithms. 

Project summary: "With the accumulation of energy data, energy domain problems are in need of more sophisticated machine learning models such as neural networks. However, building neural networks of high quality requires heavy human effort, since different hyperparameter configurations lead to significantly different performances. My project will use a multi-armed bandit approach to efficiently design the architecture of neural networks for energy domain problems such as energy demand prediction."