Imaging Science Seminar: Predicting levels of electricity consumption

Event Image
imaging science seminar sally simone fobi

Predicting levels of electricity consumption for residential buildings in Kenya from satellite imagery

Sally Simone Fobi
Ph.D. Student
Quadracci Sustainable Engineering Lab (QSEL)
Columbia University

Register Here for Zoom Link

Abstract:
Over the past decade electrification in Kenya has grown significantly, primarily due to increases in grid penetration and off grid systems. This represents a way forward for governments, international finance institutions, and entrepreneurs to increase energy access. However, electricity access planning requires an accurate estimate of electricity consumption across different locations. Electricity consumption can be inferred from various sources including direct but scarce sources like survey data and indirect but widely available sources like nighttime lights and satellite images. However, little is understood about how consumption has evolved among these newly-electrified customers. Using a unique dataset of 136k residential utility customers across Kenya over six years of electricity bills, we uncovering critical trends in the spatiotemporal evolution of electricity consumption. Our analysis reveals that recently-electrified customers are reaching their steady-state consumption more quickly than previous customers, that the steady-state is increasingly less, and that typical urban and peri-urban customers tend to consume 50% more electricity than rural customers. Next, we leverage Convolutional Neural Networks (CNNs), to predict levels of stable residential electricity consumption for buildings (upon receiving an electricity connection) using daytime satellite images. We train our model on a dataset of residential electricity bills and corresponding satellite images of the buildings. Our results show that daytime images provide more informative features and context for the task than lower resolution datasets such as VIIRS nighttime lights. While we present our results from Kenya, our method can be applied globally, thereby offering insights to decision makers for resilient planning.

Speaker Bio:
Simone Fobi is a 5th year PhD student in Mechanical Engineering in the Quadracci Sustainable Engineering Lab (QSEL) at Columbia University. Her research efforts are geared towards Electricity Demand Prediction and System Planning. She applies deep learning to detect relevant features from satellite images that can predict electricity usage for yet-to-be connected customers. This is of interest to energy providers who can leverage such models to design systems required to meet demand. Concurrently, she applies network optimization to design the placement of energy infrastructure (transformers, low and medium voltage lines), given anticipated demand. Combining both models, her research will contribute to the design and operation of more cost-effective energy systems, which offer higher levels of reliability.

Intended Audience:
Beginners, undergraduates, graduates, experts. Those with interest in the topic.
Imaging Science faculty sponsor: Dr. Jan Van Aardt


Contact
Marci Sanders-Arnett
Event Snapshot
When and Where
October 28, 2020
3:30 pm - 4:30 pm
Room/Location: See Zoom Registration Link
Who

This is an RIT Only Event

Interpreter Requested?

No

Topics
imaging science
research