The second important point which affects the performance of INNC is the number of neurons in the hidden layer.
As expected system response with the INNC of largest number of neurons in the hidden layer is better than that of small number of neurons in the hidden layer.
The following guide points are concluded form practical design, implementation and results of INNC trained by off-line training.
* The off-line training signal for the INNC must cover all ranges of the input and output variables of the system to get best training of the neural network in the off-line training phase.
* Increasing the number of neurons in the hidden layer (to certain value found by trial) that are used in the hidden layer of the INNC gives better controller performance in the real time operation.
* Good off-line training of the neural network to learn inverse dynamic of complex plant structure by using good hard training signal which is taken from the open loop response of the system, lead to good learning of the neural network and on-line training is not required in the real time operation of the INNC.
* System steady state error can be zero with the use of the INNC.
* INNC can be used for small step change in the system output.