How to train Model Reference Controller

I have the following simulink model of Active Clamp Flyback Controller and wanted to design a closed loop control but I am running into the following error when using Model reference controller:

Error using  / 
Matrix dimensions must agree.

Error in nnmodref (line 473)
step_size=ceil(max([min([(rand*(max_r_int-min_r_int)+min_r_int) max_r_int]) min_r_int])/Ts);

Error in nncontrolutil (line 19)
feval(command,varargin{:});

The inputs below generated the above error and I have no clue how the matrix dimensions are mismathcing.

I have attached both the files for reference.

Thank you.

[Sam Chak]

HI @Ajai Singh, Can you show the differential equation for the Active Clamp Flyback?

[Ajai Singh]

I do not have that, so I have the simulink model and was trying to do system identification but I was stuck there and then I closed the loop with a simple PI controller and it works. However, I need to use an optimization based controller such as an LQR or MPC but again they require a system model and that's the reason I used Neural Network based controller so that the controller can learn the dynamics and then I can tune the controller.