This repo is the code me and my team wrote for an AI hackathon with Brain and Equinor. (we were all in year 1 of our studies at this point) The task we were given was to predict the energy output of a windmill park for a 48 hour window into the future. We were given historical sensor and weather data, along with weather data and a running plan for the 48 hour window. The scoring was based on the real life earnings related to estimating power production. Predict to high and you'll have to buy a lot of energy. Predict to low and you'll have to sell your energy cheap.
To solve this task we descided to start build a simple feed forward neural network that would train on the historical data such that the weather data and running plan could be given to predict the production value of the park a given instant. Given that this was the first real AI experience for either of the team members we used a lot of time to get the basics down just to predict anything, but in the end we managed to predict comparatively with the other groups.
Although we never made any super accurate predictions we learned a lot about the basics of building and training a neural network!
I might add some details about how we solved it... but you could just read the code if you are interested ;)