Question: How to tune a model for hourly data?

[ErlendFax]

Great package, simplest I've used!

I have a dataset with hourly power consumption data for a household. As the docs say, it automatically detects the hourly data samples and adds a daily seasonality to the model. It gets the general trend pretty good as seen in the first picture. However, when looking closer it kinda misses the daily seasonality.

Shouldn't the blue line and area have a greater wave amplitude since the data has that? How can I best tune Prophet to catch this?

It does not look that bad when I zoom in, but it's still not fitting to the real consumption swings in a 24h window.

[tcuongd]

You could try to increase the Fourier order for time-of-day ("daily") seasonality (more Fourier terms = more flexible fitting): https://facebook.github.io/prophet/docs/seasonality,_holiday_effects,_and_regressors.html#fourier-order-for-seasonalities