A real-time stable neuro-controller to reduce the energy consumption in a centrifugal pump under disturbances

Abstract

In this paper, the application of an on-line tuning method based on neural networks for a PID controller was proposed to regulate the flow in a centrifugal pump. The implementation of a modified back-propagation algorithm stable in the sense of input-to-state stability was carried out to update the weights of a neural network. The energy consumed by the pump to maintain a certain flow in the pipeline of an experimental station as an indicator to assess the efficiency of the controller was chosen. Different experimental tests to show the performance of the proposed controller under different conditions were carried out such as non-disturbance, constant disturbances and time-dependent disturbances. A proportional valve was implemented to generate the disturbances in the system. The controller was compared with a classical PID controller and an on-line tuning method based on neural networks for a PID controller without back-propagation modification. The results showed that the on-line tuning method based on neural networks with a stable learning algorithm produced a lower energy consumption in the centrifugal pump.