Design of induction motor speed observer based on long short-term memory

Abstract

This paper presents a machine learning regression algorithm based on speed estimation for sensorless control of an induction motor. Long short-term memory (LSTM) based on deep learning method is used to design the induction motor speed observer. The proposed LSTM observer utilizes only the measured stator currents and voltages. It estimates the motor speed in the presence of inherent dynamics and sensor noises. Although LSTM is one of the common deep learning methods, its implementation on speed estimation for induction motor has not been tackled in the literature. The estimation performance of proposed LSTM observer (LSTMO) is investigated using four common metrics: root relative squared error, mean absolute error, mean squared error and root mean squared error. Performance of the proposed method is well guaranteed for different operating speeds. The designed observer is compared with the traditional sliding mode observer in order to prove the validity. It can be deduced from experimental results that the proposed method estimates the actual speed value successfully. LSTMO tracks the speed accurately regardless of any changes in reference speed. It is shown that there is no chattering effect on the estimated speed as compared with SMO.