I don’t understand why my predicted data (stocks) still have a gap, or a shift even after previous real data and not the predicted one. Here is the image :

X-abscissa : time (in hour)
Y-abscissa : price (in dollar)

Red curve : real prices,
Blue curve : predicted prices

Things to know,the data are used to train and to test are all normalized.
Of course I randomly mix my data.
And each row represent a series (32 quotes)

Though on the above image, prices are not normalized.
So as you can see, the price is well predicted at the beginning but the more it goes, sometimes it gets wrong, and it’s like it doesn’t take into account the real data, the corrected data.

I will provide the training file :Link to my google Drive file