When dealing with timeseries data like hourly wind data, we majorly have to deal with three types of seasonalities: daily, weekly, and yearly. Since you have mentioned you are still noticing periods even after removing daily seasonality, meaning your data still may have weekly or yearly seasonality. In order to remove them, you can follow the below-mentioned steps:
- In order to remove the weekly seasonality, you need to subtract 168 (24*7) from your data to make it weekly deseasonalized. You can do this like mentioned in the below code.
% Assuming 'data' is your wind speed time series
weekly_diff = 168; % 168 hours in a week
data_weekly_deseasonalized = data - [NaN(weekly_diff, 1); data(1:end-weekly_diff)];
- In order to remove the yearly seasonality, you need to subtract 8760 (365*24) from your data to make it yearly deseasonalized. You can do this like mentioned in the below code.
% Remove yearly seasonality
yearly_diff = 8760; % 8760 hours in a year
data_yearly_deseasonalized = data_weekly_deseasonalized - [NaN(yearly_diff, 1); data_weekly_deseasonalized(1:end-yearly_diff)];
After performing the above steps you can finally train you ARMA/ARIMA model with the deseasonalized data.
I hope it helps to proceed ahead with your workflow.
