Neural networks for optimal selection of the PID parameters and designing feedforward controller

Abstract EN

A neural network-based feedforward controller and self-tuning PID controller with optimization algorithm is presented.

The scheme of the controller is based on two unknown models that describe the system and optimization algorithm.

These models are modified Elman recurrent neural network and NARMA-L2.

The modified Elman recurrent neural network (MERNN) model and NARMA-L2 model are learned with two stages off-line and on-line, in order to guarantee that the output of the model accurately represents the actual output of the system.

The aim from the NARMA-L2 model is to find the Inverse Feedforward Controller (IFC) which controls the steadystate output of the system.

The MERNN model after being learned is called the identifier.

The feedback PID self tuning control signal for N-step ahead can be calculated the PID parameters by using the optimization algorithm with the quadratic performance index which is quadratic in the error between the desired set point and the model output, as well as quadratic of the control action.

The paper explains the algorithm for a general case, and then a specific application on non-linear dynamical plant is presented.