With the innovations every other day in the green energy sector, demands are drastically increasing and vary in intensity. The current approach to solving issues is based on models from past data. We want to bring a newer solution to our costumers. Our goal is to enable energy suppliers to optimize their energy production even in outlying cases where no model could have determined a change would have happened. This would result in both savings for the company and reduced carbon footprint through reduced energy waste.

What it does

Our app allows our clients to see their station's data and its correlation with outside elements such as the weather. Using this information, we are able to train a Recurrent Neural Network that gives us a weather forecast for the next days or months based on recent weather and live data from a Predix machine based in the each station. Secondly we predict the energy that passes through that station in that time frame based on the weather forecast. We then process the data to extract the essential information for the user, making the app accessible and easy to use for a wide range of users.

How I built it

Our app is based on: - Flask for the back-end server. - Predix for the collection of timeseries, UAA (authentification) and hosting our secured platform - Predix ui for the design of our app. - Tensorflow for the creation and training of our neural network.

Challenges I ran into

We met a few hurdles during the development of the app: - The training of a neural network of this type can be very long, adding a time constraint to an already limited time frame. - the implementation of the Intel NUC's data in our app was difficult to setup properly.

Accomplishments that I am proud of

One big accomplishment of ours is the seamless implementation of our neural network into Predix. During a previous hackathon, we had tried and failed to implement one correctly. We managed to implement a near finished application in two days, being only students with limited experience with Predix.

What I learned

During these two days of intense coding and designing, we have learned how to stream data from Predix edge to a timeseries service, implement a back-end Flask server and, most of all, optimize our time and energy to avoid losing too much of either.

What's next for PrediGE

In the future, we would like to apply our neural network to a broader dataset and for much longer in order to improve the app's precision and reliability.

Test out our app!

Use login: TEAM_6 with password TEAM_6

Built With

  • predix
  • predix-uaa
  • tensorflow
  • time-series
Share this project: