What and where should be corrected to make the forecast more or less plausible?

Question

Python

Good afternoon. I read an interesting article
https://habr.com/ru/post/495884/
I decided to use the neural network from this article for my case. In my case, the DataFrame is approximately the same as in an article with 420 thousand data, the date is used as an index. Timeframe - 5 minutes = 300 seconds = 300,000 milliseconds.
Data values ​​range from 6500-10500. I am trying to forecast data for 3 hours i.e. 36 values.
I slightly tweaked the model from the article for my case:

future_target = 36
STEP = 1

dataset = features.values
#data_mean = dataset[:TRAIN_SPLIT].mean(axis=0)
#data_std = dataset[:TRAIN_SPLIT].std(axis=0)
#dataset = (dataset-data_mean)/data_std

Until I commented out what is commented out, then my predictions varied between 1-1.2.
As a result, I get such a "beauty" with epochs and losses:

And here are the results as a result:

The question is where to look for an error and whether the model that I use is sharpened specifically for data that takes values ​​from 6500 to 10500 or after all, it is necessary to transfer data there, varying from -1 to 1?