Learning the geometry of short‐circuit faults in power systems for real‐time fault detection and classification

• Published in: IET cyber-physical systems Vol. 8; no. 4; pp. 289 - 306
• Main Authors: Naranjo Cuéllar, Juan Pablo, Ramos López, Gustavo, Giraldo Trujillo, Luis Felipe
• Format: Journal Article
• Language: English
• Published: Southampton John Wiley & Sons, Inc 01.12.2023; Wiley
• Subjects: Accuracy; Algorithms; Classification; Damage detection; Eigenvalues; Eigenvectors; Fault detection; fault diagnosis; Faults; hardware‐in‐the loop simulation; Linear algebra; Machine learning; Methods; Neural networks; Power; power grids; Programming languages; Python; Real time; Short circuits; Support vector machines; Time lag; Time measurement; Wavelet transforms
• ISSN: 2398-3396; 2398-3396
• DOI: 10.1049/cps2.12074

Given the short time intervals in which short‐circuit faults occur in a power system, a certain time delay between the moment of a fault's inception in the system to the moment in which the fault is actually detected is introduced. In this small time margin, the high amplitudes of the fault current can deal significant damage to the power system. A technique to characterise different types of short circuit faults in a power system for real‐time detection, namely AB, BC, CA, ABC, AG, BG and CG faults (and normal operation), is presented based on the geometry of the curve generated by the Clarke transform of the three‐phase voltages of the power system. The process was conducted in real time using the HIL402 system and a Raspberry Pi 3, and all programming done in the Python programming language. It was concluded that the tested types of faults can be accurately characterised using the eigenvalues and eigenvectors of the matrix that characterises an ellipse associated with each fault: eigenvalues can be used to determine the fault inception distance and eigenvectors can be used to determine the type of fault that occurred. Next, the design of a machine learning model was done based on the previously mentioned characterisation technique. The model was embedded into a Raspberry Pi 3, thus enabling fault detection and classification in a base power system in real time. Finally, the accuracy of the model was tested under different measurement conditions, yielding satisfactory results for a selected set of conditions and overcoming the shortcomings presented in the current research, which do not perform detection and classification in real time. The work proposed in this paper presents a fault characterisation technique for different types of faults based on the eigenvalues and eigenvectors of the matrix associated with the ellipse generated by the Clarke transform of the three‐phase voltages of a power system in real time.