0
$\begingroup$

I want to do forecasting for wind power generation but the problem with it is that when the wind speed is below 4 m/s the power output is zero. RNN based models do best when these type of conditions are not there. But with these consitions the RNN is not able to forecast properly. Look at the forecasting results below:

wind_forecast_RNN

As one can see most of the time I got negative results which is not possible.

I tried to using Gradient boosting regression which gave the following results.

wind_forecast_GBR

Still I'm not able to get those peak values and zero values. I don't think a single model would be able to do prediction here. So, my question is which models should I combine to get those decision making ability and regression ability?

Improve this question
$\endgroup$
8
  • $\begingroup$ Could you transform the data into integers and do count time series? $\endgroup$
    – user2974951
    Commented Dec 5, 2018 at 7:16
  • $\begingroup$ There is huge difference between data when converted into integer that is why data is nornalized between 0 to 1. $\endgroup$
    – Vedanshu
    Commented Dec 5, 2018 at 8:37
  • $\begingroup$ OK, can you bin your data into k bins between 0 and 1 after normalization? $\endgroup$
    – user2974951
    Commented Dec 5, 2018 at 8:38
  • $\begingroup$ What is k bin ? $\endgroup$
    – Vedanshu
    Commented Dec 5, 2018 at 8:40
  • $\begingroup$ Any number, say 20 bins. $\endgroup$
    – user2974951
    Commented Dec 5, 2018 at 8:40

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.