Space Apps Challenge 2017

Solarity

Our solution for the "You are my sunshine" challenge. Solarity helps you understand and see how solar power can be a benefit to you and your community. We use crowdsourced data from the ambient light sensor on smartphones to make predictions for how effective solar power will be in an area.

Process

We started by brainstorming ideas for how an application could achieve the goals set out in the challenge - to help people understand the energy output from a solar panel and plan energy consumption. We then came up with the idea of utilising the light sensor on smartphones as a means of gathering data used to make a prediction of the output of solar panels at a given location throughout the day.

Initially we planned to go with a web app, but since the HTML5 ambient light sensor API is still in draft and only supported under Firefox Mobile, we decided to go with an Android app.

The NASA dataset linked on the challenge page provides us with historical data, pre-processed with pandas, which we used to train our model on Azure Machine Learning. We then built a Flask backend to handle the API requests (to Dark Sky API and the Azure Machine Learning endpoint) and to interface with our MongoDB database to store the user data.

We then went with an Android app for the frontend. The app asks the user for their location, prompts the user to measure the light intensity by turning their phones towards the sun, and then provides the users with a chart of predicted radiation intensity throughout the day. The app also allows the user to calculate precisely how much energy they can save by installing solar panels of different types, and will provide links to more resources and our storefront (not built yet).

What's next?

The quality and variety of the dataset could be improved - we're currently only using data from the NASA dataset to make predictions, but our app itself will be part of the solution for this problem as we will be gathering the light sensor and weather data (using the Dark Sky API) every time a user makes a request. This data will be used to further refine the model.

More time could be taken to refine the machine learning model, by adding more features into the machine learning model like length of day, the precise weather conditions (clear, cloudy, rainy, etc.) which are available via the Dark Sky API but not used yet.

As for possible future steps - along with refining our MVP (minimum viable product) to something ready for release, we plan to eventually expand to new areas and expand vertically - offering additional solar panel purchasing/rental services and eventually research and production of our own solar panels, becoming a true end-to-end solar technologies company. We could give other people access to our predictive technologies and data (and also as a possible monetisation source), and expand to other uses as research, farming, health (avoiding periods of very high sun intensity), earth observation and many others.