OptCharge

OptCharge provides a machine learning solution to determine the optimal locations of new charging stations in the Ontario Essex region with the use of Google Map API, support-vector machine, and more.

Comment

Inspiration

The idea of this project came from a conversation with the non-profit Invest Windsor Essex. They were seeking a tool that strategically suggests locations for new electric vehicle chargers in the Windsor / Essex region to promote and accommodate the switch to electric vehicles

What it does

This work-in-progress project is a tool that will help them determine the most optimal locations of newly built chargers across Windsor / Essex. The best solution will take into account multiple different factors to maximize the effectiveness of this initiative. It uses a machine learning model to evaluate a location within the given region as suitable for a charger or not.

How we built it

We used a combination of data science and machine learning to build this tool with a Flutter frontend. Our backend uses a model called GPS2Vec by Yifang Yin, which enriches a location with geographical features using a deep learning model. OptCharge makes installation judgements based on those geographical features. In other words, we trained a new model to determine if it is suitable for charger or not based on the embedding vector returned by the GPS2Vec.

Challenges we ran into

Front end development was a challenge. We had to learn how to build a responsive web application with a Flutter frontend. Our team unfortunately ran out of time to fully implement the frontend. The charge hub data was a challenge. There was no provided API for data on existing charger locations. We had to use a third party API to get the data. The third party API is not free. We had to pay for it.

Accomplishments that we're proud of

Our proudest accomplishment of the hackathon has to be that we created and trained a working machine learning model that will give decision on whether a location is optimal for a new charging station. We built a support vector machine to predict the location of new chargers with SOTA embedding algorithm. We were also able to discover, comprehend, and utilize another great machine learning model (GPS2Vec) created within the last year.

What we learned

We learnt a novel way of encoding GPS coordinates into a vector space. In our partnership, it was one of our first times being exposed to machine learning models. It was a great experience figuring out how to build up the model from nothing to a functioning trained model. The frontend language Flutter is fairly new to both of us and we worked together to figure out the syntax.

What's next for OptCharge

Overall, our front-end was not completed as our team was heavily focused on the backend and training the machine learning model. Our team discussed the reality that if we had a couple weeks to complete the frontend and train the ML model with thousands of more data points, OptCharge has the potential to be a fully functional tool ready for practical use. In the future, we would like to build a mobile app that would allow users to see the location of new chargers. Also, fusing more data with the machine learning algorithm would improve the accuracy of the prediction.

Built With

Share this project:

Updates