Denoising autoencoder based topology identification in distribution systems with missing measurements

In distribution systems, the loss of existing real measurements and the stochastic penetration levels of renewable energy sources (RES) are the major issues while identifying the correct topology. The frequent switching and control actions of RES can lead to many topology changes. The loss of existi...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:International journal of electrical power & energy systems Ročník 154; s. 109464
Hlavní autori: Raghuvamsi, Y., Teeparthi, Kiran, Kosana, Vishalteja
Médium: Journal Article
Jazyk:English
Vydavateľské údaje: Elsevier Ltd 01.12.2023
Predmet:
Denoising autoencoder
Huber-loss function
Line current sensors
Missing data
Temporal convolutional network (TCN)
Topology identification
Huber-loss function
Missing data
Denoising autoencoder
Line current sensors
Temporal convolutional network (TCN)
Topology identification
ISSN:0142-0615
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:In distribution systems, the loss of existing real measurements and the stochastic penetration levels of renewable energy sources (RES) are the major issues while identifying the correct topology. The frequent switching and control actions of RES can lead to many topology changes. The loss of existing measurements from the physical meters due to failures in meters or communication aggravates the topology identification process and also it brings the convergence issues in state estimation algorithm. In this work, a deep learning approach is utilized to address the issue of incomplete measurement data and to provide accurate topology identification. A temporal convolutional denoising autoencoder (TCDAE) is developed for dimensionality reduction, reconstruction of incomplete data, and accurate topology identification. The model performance is compared with other traditional deep learning approaches such as simple denoising autoencoder (DAE), convolutional DAE, and the LSTM-based DAE. The power injection measurements and the line current sensor measurements are considered as the input measurement data for topology identification. The simulations have been carried out on IEEE 13-node and IEEE 37-node distribution test systems. The corresponding results indicate that the model can fill the incomplete data with more accurate values and its classification results prove that the model performance is effective in identifying the correct topology. •A temporal convolutional DAE is developed for reconstruction of incomplete data.•Further, the proposed model is utilized for accurate topology identification (TI).•The model performance is compared with simple DAE, CNN-DAE, and LSTMDAE.•The results show that the model can fill the lost data with more accurate values.•The classification results prove that the model performance is effective in TI.
ISSN:0142-0615
DOI:10.1016/j.ijepes.2023.109464