Power Network Topology Recognition Using Neural Networks

Author： Abd-El-Reheim Mohamed Abd-El-Aal

ISSN： 1521-0502

Source： Electric Machines and Power Systems, Vol.27, Iss.2, 1999-02, pp. : 195-208

Disclaimer: Any content in publications that violate the sovereignty, the constitution or regulations of the PRC is not accepted or approved by CNPIEC.

Previous Menu Next

Abstract

A pattern recognition application, using artificial neural networks (ANN), may be categorized as the largest category that covers the use of a neural networks for classifications. It is proved that ANNs are excellent for the applications of complex pattern recognition such as visual images of objects, speech recognition, printed or handwritten characters, and many other types of pattern recognition. Such wide applications are developed, based on the major advantage of the neural networks of being inherently error tolerant, i.e., it can solve problems adequately with information (data) that is uncertain, incomplete, or noisy. The reliable results of the power system security and economical analysis depend on the supply of real-time data base, results of power system state estimation, programs whose major problem is the determination of the network topology configuration. The determination of the topology suffers from the uncertainty of the collected data related to the circuit breakers status, due to the possible communication errors and/or failures. This problem solution matches the potential of the neural networks that can generalize, and after training, they can handle imperfect or incomplete data, providing a degree of fault tolerance. This paper describes an efficient ANN power system topology classifier. The function of such a classifier is to determine the system topology configuration under the presence of incorrect and/or incomplete data concerning the lines and the breakers status. A double layered ANN model is developed and explained as well as the specially efficient training algorithm. Comparative training and test results are reported, and they prove the robustness of this ANN.