Neural Network Controller using NARX developed in Simulink

Neural Network Controller using NARX developed in Simulink. One of the problems I encountered during Neural Network training of control appl

Neural Network Controller using NARX developed in Simulink. One of the problems I encountered during Neural Network training of control application using Simulink is the normalization issue. In this project, a simple NARX network was used to replace PID in Simulink. This is particularly useful because we addressed the following
1. How to write your own Matlab version of Mapminmax in Simulink (https://au.mathworks.com/help/deeplearning/ref/mapminmax.html)
2. The PID inputs and outputs serve as the NARX network input and target
3. Before training in Matlab, firstly normalize your inputs and target (to +1 to -1)
4. Use the "fnc" function block, replace the xmin, ymin, xmax and ymax with the maximum and minimum of your inputs and targets.
5. Since you trained using normalized inputs, you need to apply reverse normalization during the inference stage in Simulink hence the use of the third function block.