Can you make a Python program with scikit learn which adapts a deep NN model to the time series of data in time interval from 30 to 48 sec in the following csv file?  
https://techteach.no/control/python/dcmotor_v01.csv

In the file:

Time step is 0.05 s. 
# Column 0: t [s] (time in seconds) 
# Column 1: r [krpm] (disregard in modeling)
# Column 2: S [krpm] (output) 
# Column 3: u [V] (input) 
# Column 4: L_meas [V] (disregard in modeling)
# Column 5: L_estim [V] (disregard in modeling) 

Please load the data into Python workspace with numpy loadtxt, not pandas.
You can use 60 % of data for training, 20 % for testing (validation), and 20 % for recursive predictions (simulations).   

Please plot as follows: subplot(2,3,1): train S and predicted S. subplot(2,3,2): test S and predicted S. subplot(2,3,3): simulated (recursive prediction) S, and actual S. subplot(2,3,4): training input u. subplot(2,3,5): test input u. subplot(2,3,6): input u used for simulation. In all subplots: time (t) in sec should be along the abscissa axes.