Abstract

Attempts have recently been made to apply Neural Networks to control systems where they are to deal with any modeling uncertainties that may exist. This thesis proposes the Neural Network controller as a viable alternative to the conventional and widely used PI regulator for the regulation of Power Electronic converters. Neural Networks may be used to both control of and identification in a system. In general, one assumes that the mapping performed by the Neural Network can adequately represent the system's behavior over the desired operating range. PI regulators being designed for a specific load or operating point, cannot compensate for any significant change in the system parameters. This thesis presents a few applications of Neural Network control to power converters. It shows its feasibility as a current control element in dc to dc buck converters. Furthermore, the operation of an on-line Neural Network controller to waveshape the input line currents and force unity power factor operation in a voltage controlled PWM rectifier is demonstrated. Finally, for a three phase current source PWM rectifier a Neural Network controller is used to waveshape the input line currents and maintain unity power factor operation. For all three applications, this thesis presents theoretical foundations of the use of Neural Network controllers and the design considerations and guidelines for the power and control circuits. Simulation results confirm the viability of the proposed Neural Network controller and demonstrate very good performance.