power map

Inspiration

The increasing demand for sustainable and energy-efficient buildings has created a need to predict energy consumption. Georgia Tech Infrastructure & Sustainability oversees campus buildings and focuses on improving energy efficiency, yet the sheer scale of the campus causes the need for some work to be prioritized over others. Being able to conveniently quantify the energy impact of improvements would ease this prioritization process. The first step is to let building administrators examine energy data for existing buildings.

What it does

Our tool uses ML combined with weather and building data to make predictions on energy usage of buildings in America. Our solution can be applied to buildings of a variety of sizes, ages, and types. We believe our solution can significantly reduce the energy consumption of buildings and contribute to a more sustainable future.

How we built it

We trained a model using an ensemble regressor model with gradient boosting in scikit-learn and deployed it onto a Flask Server. Instead of decision trees which are one large and deep tree, ensemble models are made up of many small trees. These shallow trees are individually weak predictors but form an ensemble and collectively “vote” on a predicted value, forming a strong predictor. Gradient boosting refers to successively adding new trees to the model and calculating the importance of each tree such as to minimize the loss.

We used Google Maps API to embed a map interface onto our React website. Once the user searches for a building, essential building information is sent to a backend server, which automatically fetches any needed data from remote services to perform the energy prediction. Namely, we use open-meteo for grabbing weather data, and Precisely for grabbing building information.

Challenges we ran into

There were only two of us throughout the entire project. We have never had hackathon experience beforehand. A major obstacle was working with Google Map API, which provides a complex but also a deep set of features.

Accomplishments that we're proud of

We made our first website ever in production, and on GCP no less! One of us is a freshman who learned a lot during the process - React and APIs were used for the first time in a real project.

What we learned

Because this is our first hackathon, we learned how to iterate quickly and brainstorm new ideas in a short period. We learned a lot about doing data visualization and effectively using geographical tools to improve user experiences. We also gained experience in methods of deploying ML models in production.

What's next for power map

This hackathon project will be presented to Georgia Tech Infrastructure & Sustainability. We'll collect feedback from stakeholders on the user interface and features that need to be added. We'll also improve the ML model by evaluating different imputation techniques, which can help deal with missing data.