Simulation-driven deep learning for locating faulty insulators in a power line

Blazhe Gjorgiev, Laya Das, Seline Merkel, Martina Rohrer, Etienne Auger and Giovanni Sansavini

Reliability Engineering and System Safety, 2023, vol. 231, issue C

Abstract: Overhead power transmission lines are repeatedly subject to environmental fluctuations that affect their insulation. Weakened insulation allows leakage current to increase over time, eventually triggering the protection and thereby causing unplanned power line outages. Current advancements in sensor technologies, data analysis, and machine learning make it possible to develop tools for online monitoring of power lines. The availability of data corresponding to different operating conditions plays an important role in the development of such tools. However, obtaining an adequate amount of data for different types, severity, and location of faults to reliably identify distinguishing patterns may not be feasible. Under these circumstances, a physics-based model of the line allows one to generate data for different insulation degradation states. In this work, we couple a machine learning approaches with a physics-based model of an existing power line in the Swiss power grid. The physics-based model is calibrated to produce leakage currents that are well aligned with measurements. The model is then used to produce large sets of leakage current data while simulating different states of fault at different locations along the power line. The faulty states are generated by varying the capacitance at different segments of the line. The generated data form the basis to train multiple state-of-the art machine learning methods, including long short-term memory and convolutional neural networks. Furthermore, we exploit different extraction techniques and automate the machine learning pipeline using custom layers. The objectives of the learned models are to detect the existence of a faulty insulators and their locations in the power line. The obtained models show accuracy higher than 99% on a test data set, which represents smaller part of the generated data set that has not been used in the training of the models.

Keywords: Physics-based model; Leakage current; Power line; Machine learning; Deep learning (search for similar items in EconPapers)
Date: 2023
References: View references in EconPapers View complete reference list from CitEc
Citations: View citations in EconPapers (1)

Downloads: (external link)
http://www.sciencedirect.com/science/article/pii/S0951832022006044
Full text for ScienceDirect subscribers only

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:eee:reensy:v:231:y:2023:i:c:s0951832022006044

DOI: 10.1016/j.ress.2022.108989

Access Statistics for this article

Reliability Engineering and System Safety is currently edited by Carlos Guedes Soares

More articles in Reliability Engineering and System Safety from Elsevier
Bibliographic data for series maintained by Catherine Liu ().