CONTINUOUS METHOD FOR LEARNING TIME SERIES DATA

We use a chain to learn time series data.
For example, create a model by repeatedly learning EPOCH:100 with data from A1 to A99.
We expect the following data (assuming Q100):
Save the model and reflect the difference between the actual A100 and the Q100 in the model through the loss function.
Expect Q101.

Well, that's all right now.
When learning the difference between Q100 and A100, I am having trouble understanding how to repeat the EPOCH number of saved models.

Since it is time series data, I don't think it's enough to repeat the difference between A100 and Q100 100 times.
It's a model that I've learned to some extent, so I think it's better to let them learn without repeating it.
Wouldn't it have a negative impact on your expectation if you didn't repeat the A200 or A300 before you went ahead?
It is questionable that
If I want to relearn, do I need to repeat A1 to the latest data?

If anyone knows, could you please let me know?
Thank you for your cooperation.