Space Apps Challenge 2017

Background

We addressed the challenge of making people understand the energy output of the their solar panel. This entails creating a medium to explain the factors that are responsible for the generation of electric energy from solar energy via the solar panel.

We also looked into helping people get a good idea of the amount of electric that should be produced from their solar panel the next day so as to aid planning of their energy consumption. We ensured it was simple for the layman to understand and also made provisions for HI-SEAS personnels and other space explorers.

Resources

• YouTube Video: We embedded a video which clearly educates on the energy generated from a solar panel
• We used https://freegeoip.net api for client location detection.
• We used open weather map api for weather conditions detection based on location.
• Weather data provided in csv format by NASA.

Technical Approach:

There are three independent variables required to estimate the total power output from a solar cell. The variables are total solar radiation per area given in watts per metre, module efficiency of the solar panel given in percentage and the total exposed area of the solar panel.

Most of the available online tools usually require some complex input from users. Which is because they actually need them to estimate this values. Unfortunately end users who are not very familiar with this terminologies would not be able to use this tools. We decided to solve that problem. A tool which would be available for technical and non-technical users.

In order to estimate the solar radiations, we used a learning algorithm based on the weather data provided by NASA. The data contains humidity, Barometric pressure readings, wind speed, wind direction, sunrise, sunset and solar radiation per area. There were approximately 33,000 rows of data. We had to drop sunrise and sunset data because number of rows of data that was unavailable was about 700. Using pandas framework in Python, jupyter notebook, numpy, scikit learn, xgboost et.c, we wrote a linear regression algorithm that gave an accuracy of about 70%. The model was then saved.

To interact with the user through an API (which would allow for mobile interface) and a web interface, we proceeded to build the backend with flask, a python web framework. We used flask to build a public api which if given the necessary parameter would output solar radiations. Some of the parameters include barometric, humidity in fraction, wind direction, wind speed, module efficiency, and exposed area.

The front end was built with the goal of collecting user data in the most friendly way.

We first required our user to indicate their location. If they were on earth, it means we could automatically determine their current latitude and longitude using https://freegeoip.net api else we assume they were on a spacecraft and would be able to easily read all the data from there. For those on earth, we could also automatically determine their humidity, wind speed, wind direction and barometric pressure readings using open weather map api. Fig 3 shows how we estimate efficiency by asking our users for their panel manufactures which they can easily check up and also asked them to estimate the size by comparing their panel with a picture. With all these, we could determine their total ratings.

Finally, what they know is how much their Solar panel can generate in “Watts” (i.e energy per hour). This probably doesn’t make sense to non-technical users. What they want to know is how many of their appliances can they use with it. Fig 4 is our final page a planning tools that already knows the user’s total panel ratings. It displays common household appliances and their estimated power ratings in watts. The user can add any appliance, increase and reduce it’s value. The chosen appliances are listed and then when the total power ratings is above the total panel ratings, the value turns red in order to warn the users. The space option was created with NASA’s HISEAS team in mind.

Fig 1. The welcome page where we require user location

Fig 2 The page created for space explorer’s input.

Fig 3 The page for non-technical user’s input

Fig 4 The page where the non-technical user can check which appliances can be used with the solar panel based on the expected output.

Challenges Faced

• We successfully hosted to heroku, But the app could not work well because one of the open weather map api is configured to load data over http and since heroku does not permit that, there was an error. This could easily be solved by hosting on a digital ocean or AWS(Amazon web Services) server for example.
• We had to drop sunrise and sunset data because number of rows of data that was unavailable was about 700. Going ahead to use it like that would affect in a great proportion the reliability of our prediction.

Prospects

• Mobile app
• Sms api for local users
• Using convolutional neural network, we could create an algorithm that learns to recognize some visual distinctions in solar panels to help estimate efficiency and size. So all the user has to do is use their mobile phone to take a picture of the solar panel and they would know the ratings.
• A similar algorithm could be used to estimate power ratings of common household appliances by just taking pictures with mobile phones.

Conclusion

Beyond the fact that this is a competition, we understand how machine learning is disrupting technology in our current generation. We can easily link up to NASA because of the large amount of useful dataset available. With products like this, we believe this large amount of available data is not only going to help us solve some of the greatest problems in our nations and in the world but it can inspire a whole generation of great thinkers and innovators. Like Deepmind, we can solve intelligence and use it to solve everything else, Like google, we can make all this accessible information useful and eventually benefit mankind by revealing the unknown or esoteric knowledge.