Research On Parameter Identification Of Asynchronous Motor Based On Recurrent Neural Network With Short And Long Memory Patterns (LSTM-RNN)

Induction motors are widely applied in modem industry due to their excellent features such as simple structure,low manufacturing cost,small quality,convenient operation and maintenance,and reliable operation.The appearance and improvement of induction motor vector control improve the motor driving performance.As a key part of vector control,the accuracy of parameter identification affects the performance of vector control,especially the accuracy of the rotor resistance.The parameter identification of induction motor is divided into off-line identification and online identification.The off-line identification can provide relatively accurate initial parameters of the motor parameters,but it has no significance for the motor in operation.Therefore,it is necessaiy to deal with the adaptive parameter identification so as to make the asynchronous motor speed control system run efficiently and steadily.Rotor resistance identification value is influenced by many factors and parameters are easy to change,need to solve for online identification,parameter identification model of asynchronous motor classic low accuracy and easy to fall into local optimal solution and over fitting problems in the parameter identification,this paper proposed the LSTM-RNN(long short term memory recurrent neural network)a new method of induction motor rotor resistance online identification of the length of the memory type neural network.The results of LSTM-RNN neural network identification are compared with that of BP and RNN.The experimental results verify the validity and correctness of the proposed method.The main contents of the full text are as follows:1.The research status of asynchronous motor parameter identification and the research status of induction motor parameter identification based on neural network and the research status of RNN/LSTM are introduced.2.The dynamic mathematical model of asynchronous motor in three coordinate systems is introduced,and the expressions of the three coordinate systems are derived.3.The structure and learning method,training process and research theory of BP and RNN neural network are introduced.The research method of rotor resistance parameter identification of asynchronous motor based on BP and RNN neural network is expounded,and the simulation results are analyzed and compared.4.The theoretical knowledge and model structure of LSTM-RNN long and short memory neural network are studied.The LSTM-RNN neural network is used to iderntify the rotor resistance of asynchronous motor online,and the siinulation results are analyzed and compared,which will prove the superiority of the new neural network.5.The online identification of rotor resistance is done by using the automatic encoders and noise reduction encoders in deep learning.Simulation and experimental results verify the effectiveness of the method.6.The experimental platform for parameter identification of asynchronous motor is set up to collect the data.The experimental results verify the validity and correctness of the above method.