Application of adaptive neural network control algorithms for asynchronous electric drive to optimize milking umit operation modes and minimize udder trauma

The article addresses the problem of ensuring the stability of the working vacuum in milking units. It is shown that dynamic vacuum fluctuations, caused by unpredictable changes in technological load, are the main reason for udder trauma and the inefficiency of classic PID controllers. The aim of the work is to develop and substantiate a predictive control system for the frequency of an asynchronous electric drive based on an LSTM neural network controller. Unlike reactive PID controllers, the developed method uses the LSTM module for anticipatory forecasting of vacuum dynamics and forming a corrective control signal. Simulation was carried out in the MATLAB/Simulink environment under a critical step load drop. Numerical results confirmed the high efficiency of the predictive approach: the maximum vacuum deviation (𝛥𝑃𝑚𝑎𝑥) was reduced by 81.5% (from 6.5 kPa to 1.2 kPa), and the root-mean-square deviation (𝜎𝑝) was reduced by 77.3%. The settling time was reduced by almost three times. The implementation of predictive control guarantees the maintenance of vacuum within the safe technological corridor (50±2 kPa), eliminating udder trauma, and demonstrates a reduction in electric drive energy consumption by up to 4.5% in dynamic operating modes.