Neural network input / NN Predictive controller

I am using a neural network in a heating system, trying to train the system. I ran the first time the system turns on the Heating with a PWM from 0° to 120° in 2500 seconds and then stays constant around 120. I took these data (X=Time/Y=Temperature) I also took the PWM figure data (X=Time/Y=%of the PWM). As I am using the NN Predictive Controller in SIMULINK, I would like to ask if the input data will be the temperature that I previously took from the system, and the output data will be from the PWM readings? Last question, I saw many articles mentioning that inputs have to be normalized, how about Target data?