The Research Of Induction Motor Vector Control System Based On Neural Network

In recent years, induction motor has been extensively used in motion-control field with the advances in power devices, micro-processor and motor control theory. At the same time, its high performance is required increasingly. On one hand, as the performance of induction motor vector control system will be influenced by parameter variations, external load disturbances, etc., the high performance requirement can't be met simply using traditional control methods. To make the system be robust to the influences, new control strategies have to be developed. On the other hand, the mounting of speed sensors brings some disadvantages to the induction motor vector control system in which rotor speed is closed-loop controlled. Therefore, the research to speed-sensorless drive system and enhancing its performance has become a popular research aspect in AC drive field. Considering the superior performance in the control and identification to nonlinear, complex and uncertain systems, neural networks was used in induction motor vector control system combined with PI control and model reference adaptive control in this dissertation, which has improved system performance.At first, in order to solve the problem that satisfactory performance can't be achieved using conventional PI controller when the system parameters vary in a wide range and reduce the influence of parameter variations and external load disturbances, the adaptive neural network PI speed controller and neural network model reference compensation controller were used in induction motor vector control system in the dissertation. The controller was constructed according to the characteristic of neural networks and vector control system and combining neural network with PI control and model reference adaptive control. Then, adaptive control to motor parameters and the load torque was realized.In the application of the neural network model reference compensation controller, to enhance the mapping capability and dynamic response capability of the system, a three-layer forward neural network was used in the controller and adaptive control algorithm with generalized PID was used to correct the weights of the network. In a certain extent, the speed of convergence was accelerated by using the algorithm.Subsequently, the flux observation method, stator resistance on-line identification method and speed estimation method based on the neural network PI controller were presented in the speed-sensorless vector control system.In the rotor flux-oriented system, in order to guarantee the effective decoupling of vector control, the precise flux orientation is needed. Among many flux observation methods, voltage model and current model have their own advantages and disadvantages, and they are complementary in many ways, so they were combined in many flux observers. A closed-loop flux observation method using voltage-current combination model based on the neural network PI controller was presented in the dissertation, which realized the rotor flux observation and solved the instability of voltage model through the current model's adaptive correction to the voltage model in stationary axes,.In closed-loop flux observer using combination model, the stator resistance's error by the temperature's variation will impact on the accuracy of rotor flux observer and speed control, destroy the performance of the system. Therefore stator resistance on-line identification method based on neural network PI controller was presented, by which the accuracy of speed control was improved.About the core of the speed-sensorless vector control system, that is speed estimation, in order to improve the performance of the system, in which the speed signal is structured with PI control, a speed estimation method based on the neural network PI controller was presented. In the method, the PI parameters were set as the weights of the neural network. When the input error varied, the weights were adaptively regulated, by which the convergence rate was accelerated, at the same time, the overshoot and stabilization time were reduced.Finally, the experiments were done in the induction motor vector control system platform based on TMS320F2808. In the system with speed sensor, neural network PI controller was used for speed control. The experiment results verify the program is valid and feasible. In the system without speed sensor, the estimated speed is closed-loop controlled. The experiment results verify that system's dynamic performance can be improved by using the speed estimation method based on neural network PI controller presented in the dissertation, the stability of the system was guaranteed in sudden increase in load disturbance and low-speed operation, the current waveform and system performance can be improved effectively by using the dead zone inverter voltage compensation method.