Research And Improvement On Speed Identification Based On Neural Network MRAS

The high-performance vector control system needs to control the speed with closed-loop strategy, but in many cases the use of the speed encoder disk and other speed measuring device is limited. So the study of the speed sensorless vector control system has an important practical significance. Speed identification is an important part of speed sensorless vector control system, the accuracy of speed identification has a direct bearing on the overall performance of the vector control system. Due to the own deficiencies of PI adaptive law, the traditional model reference adaptive speed identification produces the large speed identification error and the low recognition accuracy. At the same time, changes in the motor parameters can also seriously affect the effect of the speed identification.In response to the above questions, this thesis proposes the method of neural network MRAS speed identification on the basis of the model reference adaptive speed identification. This method uses BP neural network algorithm instead of the PI adaptive institutions of model reference adaptive speed identification, so as to effectively overcome the deficiency of PI adaptive law and get the high precision of speed identification. In order to reduce the effect of changes of the motor parameters on the speed identification, the MRAS rotor time constant online identification is proposed and speed sensorless vector control system with rotor time constant online identification is designed.Firstly, this thesis compares the two kinds of speed estimation methods by simulation studies, and the results show that the neural network MRAS speed identification has faster convergence rate and higher identification accuracy. Secondly, through simulation, this thesis analyzes the effect of the changes of rotor time constant on the neural network MRAS speed identification, and verifies the correctness of the rotor time constant identification based on MRAS. Finally, the speed sensorless vector control system with rotor time constant identification is studied by simulation. Simulation results show that the online identification with rotor time constant can be effectively reduced the impact of the changes of rotor time constant and improve the accuracy of the speed of identification.Based on the TMS320F2808DSP core of the company of TI, this thesis designs the hardware circuit, including the main circuit, switching power supply circuit, drive circuit, detection circuit, protection circuit, etc, and completes the preliminary software design.