Auto-regressive models effectively capture typical time series behavior where time steps that are close together are highly correlated. Here, an probabilistic auto-regressive model simulates hourly ERA-5 wind speed and direction data which have a high degree of auto-correlation.
However, unlike simple auto-regressive models, this simulation mimics the long-term seasonal and daily trends of the training data set using a fourier fit.
Look below to see the local phenomena generated at a time-step by time-step level as well as the long-term features generated such as distribution, wind rose, and monthly/dirunal profiles.
The example can generate any number of time series to mimic the original ERA-5 time series provided, and the approach is generalized so that it can match any wind speed/direction training set that you feed to the model.
| Original | Generated |
|---|---|
 |
 |
 |
 |
 |
 |
 |
 |