I'm designing a neuro-controller for a complex dynamics system. the plant that i'll have to control is
-9th order dynamics system
-2 inputs 2 outputs plant
I try to train the neuro-controller by using specialized training of direct inverse control with a reference model to make target output response less aggressive. i trained it in batch mode using MATLAB commands available. however, i found it difficult to converge. i use Levenberg-Marquardt algorithm but it reached maximum MU easily and training halted, although the neuro-controller isn't able to control the plant yet.
Does anyone know what problem it might be?