Research Of The Theory Of Neaural Network Robust Adaptive Control Of Speed-Sensorless Induction Motors

This thesis is focused on the speed-sensorless induction motor control system. Some new approaches are proposed for four aspects such as rotor flux estimation, rotor speed estimation, neural network robust adaptive control for induction motor and its other applications.For the condition that the rotor resistance and load torque are unknown, a sliding mode variable structure approach for rotor flux estimation is proposed. Then its improved algorithm is proposed to restrain the DC disturbances. The results of simulation indicate that this approach has high precision.An observer that can estimate the rotor flux and speed is proposed. Its algorithm is proposed too. The results of simulation indicate that using this observer the stator resistance, rotor flux and speed estimation have high precision. For the condition that the rotor resistance and load torque are unknown, using the voltage model to estimate the rotor flux, a speed observer that need estimated rotor resistance is proposed. The rotor resistance is estimated through using Popov hyperstable theory and injecting sinusoidal AC(Altering Current) signal to rotor flux reference signal. The reason of the need of injection signal is analyzed. The results of simulation indicate that the precision of the estimation of the rotor resistance and speed is high. To improve the precision of speed, a new approach that can calculate the rotor resistance and speed directly is proposed. The rotor flux is estimated using voltage model. Then, the equations of rotor resistance and speed are deduced from the stator frame model. The expressions of speed and rotor resistance are solved from those two equations. The condition that the denominators of them are zeros is analyzed and the way to deal with that condition is presented. The results of simulation indicate that the proposed approach can estimate the rotor resistance and speed with high precision.To solve the problems of the control of induction motor with unknown rotor resistance and load torque, the L2-gain robust controls are designed using backstepping without solving the complex HJI inequality. To avoid the explode of terms caused by backstepping, the dynamic surface control technique is adopted. The robust terms are introduced in the outerloops to restrain the affect of uncertainties. The control system is proved to have the L2-gain that is less than or equal to the The control system is proved to have the L2-gain that is less than or equal to the positive constantγusing Lyapunov theory and HJI inequality. The simulation results indicate that the tracking and robust performance of the proposed approach is higher than the system without the robust terms.A neural network L2-gain robust adaptive control approach for induction motor is proposed in this thesis. In this control approach, the RBF neural networks are adopted to compensate the uncertainties caused of rotor resistance and load torque. Based on backstepping, the controllers are designed without solving the HJI inequality directly. The projection algorithm is presented as the learning law of the weights of the RBF neural networks. The control system is proved to has the L2-gain that is less than or equal to the positive constantγusing Lyapunov theory and HJI inequality. The results of simulation indicate that the proposed method has high performance and strong robustness. On the other hand, it has the situation that the system is out of control. Then in that case, an improved neural network L2-gain robust adaptive control approach is proposed. The controllers are deduced using backstepping, without solving the HJI inequality. The disturbances caused by the unknown rotor resistance and load torque are considered, and the RBF neural networks are applied to compensate those disturbances. Then the tracking errors of the state variables of the rotor flux oriented coordinate system rotating two phase model and the weights of the RBF neural networks are taken as the state variables of the whole control systems. The improved learning algorithm of the weights is proposed which is a kind ofδ-modify algorithm. The zoom in factors are introduced in the output expressions of the RBF neural networks to reduce the evaluation signal norm of the whole control system. The whole system is proved to be stable using Lyapunov theory and HJI inequality. These two control methods are both applied in company with the third speed estimation method proposed in this thesis. The simulation results indicate that the proposed approach has high dynamic performance and is robust to the considered uncertainties.A neural network robust adaptive control scheme for induction motor is proposed in order to deal with the uncertainties of stator resistance and load torque. Using backstepping, the controllers that can limit the state variables and the weights of the neural networks in the prescribed range are designed. The corresponding learning law that is another kind ofδ-modify algorithm of the neural networks is proposed. The stableness is proved using Lyapunov theory. This control method is The simulation results indicate that the proposed scheme has high tracking performance and strong robustness.As a typical example of second-order DC/DC converters that has the nonminimum phase characteristics, the average model of Buck-Boost converter is analyzed. The average model of the Buck-Boost converter is transformed to be a first order uncertain nonlinear system using a nonlinear feedback that is taken as the inner loop controller. Employing a RBF neural network compensator, a neural network robust controller is proposed as the outer loop controller to control the output voltage of the Buck-Boost converter. The tracking errors and the weights of the neural network are proved to be ultimate consistently bounded using Lyapunov theory. The results of simulation indicate that the controller has high dynamic performance and is robust to the considered uncertainties of the converter. For a class of uncertain triangular structure nonlinear SISO systems, the controllers that apply the RBF neural networks to compensate the uncertainties of the controlled system are designed without solving the HJI inequality using backstepping and dynamic surface control technique. The performance indicator of L2-gain is chosen. The tracking errors of the states of the controlled system and the weights of the neural networks are taken as the state variables of the whole control system. The improvedδ-modify algorithm is presented. The zoom in factors are proposed to reduce the evaluation signal norm of the whole control system. The Lyapunov theory and HJI inequality is applied to prove that the proposed controllers are the neural network L2-gain robust adaptive controllers of this class of system. The results of simulation indicate that the proposed approach has high tracking performance and strong robustness.In the end, the main innovations of this thesis are summarized, and the fields of future investigation are expected.