Insample forecast of Auto-SARIMA very good, forecast of SARIMA Learner/Predictor with Auto-SARIMA hyperparameters poor

Hello KNIMERs,

As you can read in the title, I have a time series on energy consumption in hours with two seasonalities: 24 and 168. If I run Auto-SARIMA with the first 500 values, I get a very good quality insample forecast (see Auto-SARIMA Summary). I use the hyperparameters obtained in this way in the ARIMA Learner and let the SARIMA Predictor predict the next 200 values. In the line plot, however, one can see that the forecast does not turn out well or does not work out the second seasonality. Does anyone have any ideas?

Thanks a lot
Auto-SARIMA example.knwf (777.0 KB)

Auto-SARIMA workflow

Auto-SARIMA Summary

Signal vs. prediction

This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.