How to turn a custom SNN into a good controller ?

Hello, I've learned how to create a completely custom neural network in .m file and exported it into simulink by following a tutorial on Mathworks.

Now I'd like to use it as a controller for an nonlinear plant. It's the most simple closed control loop "input->error->SNN controller->plant->output" and while the simulation executes and there are no errors, the quality of control is poor to put it gently.

I don't really know how to teach an network to be a good controller. I've simply used 'plant's output' as 'SNN input' and 'plant's input' for 'SSN output' (pretty much inversed the plant).

How can I make SNN into a decent controller ?