Using LSTM in a model?

Question

Python

I am modeling the parameters of a gas turbine engine using a neural network with LSTM layers. I get a fairly accurate model, with the exception of falling down and up at the "thresholds" of values. Tried various combinations of activation functions, optimizers, and number of layers in the model, but couldn't solve the problem. Please tell me what I am missing or doing wrong.

My code:


The orange line is the original data and the blue line is the simulated data:

Answer 1

It looks like the original function is not smooth, more like discrete numbers. This means that the smooth function at the output is trying to model a discrete discontinuous function of the initial data - of course, it will wedge at the discontinuities. Can smoothing what for initial data to add?