Research On Direct Torque Control For Induction Motor Based On Neural Networks

At present, the direct torque control technology is widely used in AC variablespeed system, because it has many advantages such as simple control structure, fasttorque response, stronger robustness and so on. Then, the problem of complex compu-tation structure and sensitive to parameter variations in vector control to be solved.Butdirect torque control technology still has some shortcomings that need improvement.The accuracy of the stator flux observer and speed observer, which affect the perform-ance of direct torque control. Based on deeply and extensively study on the veracity of statorflux observer and speed observer, then used neural network theory, a new stator fluxobserver method and speed observer method is proposed in this dissertation. The mainresearch is as follows:At first, three conventional flux observers is described in this paper, and then theneural network theory is to be introduction. For achieving the accurate observation ofthe motor flux, a closed-loop flux observer with adaptive capacity is proposed, whichis consists of the actual motor using as a reference model and the adjustable systemthat the motor flux model and BP neural network inverse flux model constitute. Byadaptive controlling the inverse flux model output is made to track motor output, sothat the output of the flux model can approach the actual motor flux. So this methodhas a stronger adaptive ability and good robustness,it can be better estimated flux thanthe open-loop flux observer when the motor parameters is to be changed.Because of neural network have a better of generalization ability, adaptive abilityand nonlinear mapping ability, so the neural network speed observer program isproposed in this paper. But the randomness of choice of the initial weights andthresholds, it can not be accurately obtained and impact on BP neural networkperformance. In order to improve the speed of identify effect, the combination of thePSO algorithm and BP algorithm in this paper, then to obtain the best initial weightsand thresholds by this method. Used the experimental data collected to train the neuralnetwork and experimental results demonstrate the effectiveness of the method.This neural network closed-loop flux observer is applied to the direct torquecontrol system and the simulation model of proposed design schemes are built up withMATLAB7.10/Simulink, then it compared with conventional direct torque controlsystem. Computer simulation results demonstrate direct torque control system ofneural networks is not sensitive to the motor parameter variations and load disturbance, and has a good dynamic and static performance.