Neural network tuning for PID controller

Hello, I have a problem with neural networks. I want to create one, that tuning PID-controller (outputs of NN are integral, proportinal and derivative coefficients). In perfect case this NN should work in simulink model and training during simulation. It means, that weights changes during simulation and coefficients do the same. Inputs of NN not important. The main question is how to train neural network during simulation of SIMULINK model and is it possible at all? Or maybe you have some advice how to do it in different way. Thank you in advance!

[viyils Sangregorio]

You were able to sole your poblem ??