Wind Turbine Predictive and Uncertainty Analytics

What it does

Currently, not much.

How I built it

I spent most of the time trying to build new features to find if I could improve the prediction of the nacelle weights and the turbine blade weight. In the end, looking at the features that had high independence and had the most influence in the XGBoost model that was first used to fit the data only a couple of the parameters actually mattered in fitting it and so the final model just uses the turbine power rating, a parameter to do with the blade geometry called a1 and the water depth that the turbine was in. This model was still really awful and still has errors of around 40/50% which is probably not much better than just guessing.

Challenges I ran into

The dataset was small and sparse. I also ran out of time.

What's next for Wind Turbine Predictive and Uncertainty Analytics

Given more time I would like to have developed a way to generate more training data or to have done some web scraping as the dataset I had was not big enough to give me good results with the tools I know how to use. I also didn't quite finish the challenge as I don't have a confidence interval for my values but trust me when I say that those would be awful too.

Feel free to have a look on Kaggle at how it went.

Built With

• kaggle
• python