Identifying topology in power networks in the absence of breaker status sensor signals

• Published in: IEEE Mediterranean Electrotechnical Conference pp. 160 - 165
• Main Authors: Oliveira, Rui, Bessa, Ricardo, Iranda, Vladimiro M
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.05.2018
• Subjects: Artificial neural networks; deep learning; Entropy; information entropy; Load flow; Network topology; Topology; topology estimation; Training
• ISSN: 2158-8481
• DOI: 10.1109/MELCON.2018.8379086

This paper presents the concept of a tapered deep neural network, subject to unsupervised training layer by layer, under a criterion of maximum entropy, to perform the estimation of breaker status in the absence of a specific sensor signal. The almost perfect prediction power of the model confirms the conjecture that the knowledge of the topology of a network is hidden in the electric measurement values in the network. A test case is presented with computing speed accelerated by using a GPU (graphics processing unit). The comparison with a previous model illustrates the superiority of the novel approach.