Adaptive PID Neuro-Controller for a Nonlinear Servomechanism

In this paper we propose an adaptive PID control scheme based on recurrent neural networks (RNN). The control system includes a RNN-PID control network and a RNN emulator. The recurrent neural networks are trained on-line using the RTRL learning algorithm. The plant sensitivity Information is calculated on-line using the emulator network and is fed back along with other inputs to train the control network. On-line simulation studies and results for a one-degree of freedom robot arm servomechanism are presented to show the effectiveness of the proposed control scheme.