Application of NARX Artificial Neural Networks for Pressure Identification of Liquid Transfer Station with Asynchronous Electric Drive

The article presents a methodology for developing a pressure neural identifier for a liquid pumping station with an asynchronous electric drive. The widespread use of these units in industry and the housing and utilities sector determines the relevance of considering these issues. Increasing the level of automation of these units is possible through the use of additional feedback on the process parameter - pressure. Pressure is calculated using a neural network - autoregressive and exogenous inputs. Pressure is determined by using information from alternating current and voltage sensors of an asynchronous electric motor. In addition to these parameters, the amplitude of current and voltage, as well as the obtained pressure with some time delays, are used to expand the vector of input coordinates. The work presents a simulation model of the electric drive of a pumping unit, used to generate input data, as well as to evaluate the effectiveness of the applied structure and training results. The integral criterion of the minimum difference was used as this assessment. The resulting identifier is stable in static and dynamic operating modes.