Neural Network NARX Flow Identifier for a Pumping Unit With an Induction Motor

The article presents a description of the operation of a neural network identifier for determining fluid flow in centrifugal pumping units. We are considering a fluid transfer station, which includes an inlet and outlet pipe from a centrifugal pump, suction and pressure pipelines with valves, as well as an asynchronous electric drive with the ability to adjust the frequency. In modern pumping installations, this type of electric drive is often used in industry. Improving the quality of operation of liquid pumping stations is associated with improving the quality of supply regulation. For this purpose, feedback based on liquid flow meters is used. The article describes the experience of obtaining a volume liquid supply signal based on a measured input signal vector consisting of voltage and current signals supplied to an electric motor. During the identification process, the vector of input coordinates is processed using a nonlinear autoregressive neural network (NARX). For this purpose it is pre-trained. The use of the NARX network structure made it possible to increase the dimension of the input vector, which led to increased identification accuracy and operational stability.