Artificial neural networks for state estimation of electric power systems

This thesis deals with the application of Artificial Neural Network (ANN) technology in power system state estimation. It addresses the following developments: the fundamentals of the state estimation based on ANN technology; the feasible ANN state estimation methods; use of voltage phasor angle measurements in ANN state estimation; and bad data processing for ANN state estimation. The power system state estimation problem is formulated as an optimization problem applied to dynamic ANN model. Two types of dynamic ANN models are used: ANN model with steepest descent dynamic; and ANN model with Hopfield-style dynamic. The complexity of an ANN State Estimator (ANN SE) is reduced by using the following techniques: a special algebraic transformation of the ANN objective function; and the incorporation of zero-injection measurements by the using variable reduction technique. At the same time, these two techniques improve the filtering performance of the ANN SE. Two methods for designing the ANN SE for a specific power system are developed: an analytical method: it maps the structure and the parameters of a power system into the ANN SE structure and parameters; and a synthetic method: it is based on the Real Time Recurrent Learning (RTRL) technique (used in training dynamic ANN), where the ANN SE structure and parameters are determined through learning from available input/output (measurements/state variables) data. In continuation of the thesis feasible ANN SE methods are developed.

Grantor dc:publisher.institution

Department of Electrical Engineering

Year dc:date.issued

1996

Author dc:creator

• Zivanovic, Rastko

Advisor dc:contributor.advisor

• Petroianu, Alexander

Language dc:language.iso

eng

Handle dc:identifier.uri

http://hdl.handle.net/11427/9470

OAI identifier oai:identifier

oai:open.uct.ac.za:11427/9470