Application of neural networks for classification and system identification in power systems

This dissertation investigates the application of layered perceptron neural networks in security assessment and high performance dc motor control. Security assessment is investigated both from a static and dynamic security perspective. Neural network classifiers are trained to identify the unknown underlying characteristics between measurable power system variables and the post-contingency security status.;Static security classification is based upon pre-contingency steady state variables while dynamic security classification is based upon transient variables measured at fault clearance. Statistical feature selection techniques are used to reduce the dimensionality of the input space while enhancing classification accuracy. The motivation is to efficiently compress the input space so that large scale power systems can be handled with relatively small neural networks. Training is achieved by learning a set of input/output examples through back error propagating.;The ability of a layered perceptron to identify the forward and inverse dynamics of a dc motor and use the acquired knowledge in the trajectory control of motor speed and/or shaft position is investigated. The noise rejection and knowledge generalization capabilities of the layered perceptron are effectively used in order to achieve a robust controller design applicable to a wide range of operating conditions.;Different neural net topologies that can capture the forward and inverse dynamics of a dc motor are designed and evaluated through simulations. The main design criteria is to achieve trajectory control of motor speed and/or shaft position through the concepts of model reference adaptive control (MRAC).;The control algorithm is implemented on a laboratory hardware setup. The neural network controller is assembled within a PC-based, real-time, control system shell, using software subroutines. A computer controlled, dc/dc, voltage converter is used to generate the specified control signal sequences for driving the motor. Different test conditions are used to establish the robustness, noise immunity, and stability of the controller. (Abstract shortened with permission of author.)...