Data-driven failure diagnosis in transmission protection system with multiple events and data anomalies

Abstract

To guarantee the reliable power supply, the expected operation of all the components in the power system is critical. Distance protection system is primarily responsible of isolating the faulty section from the healthy part of the grid. Failure in protection devices can result in multiple conflicting alarms at the power grid operation center and complex events analysis to manually find the root cause of the observed system state. If not handled in time, it may lead to the propagation of the faults/failures to the adjacent transmission lines and components. With availability of the synchronized measurements from phasor measurement units (PMUs), real-time system monitoring and automated failure diagnosis is feasible. With multiple adverse events and possible data anomalies, the complexity of the problem will be escalated. In this paper, a PMU based algorithm is presented and discussed to detect the root cause of the failure in transmission protection system based on the observed state, e.g. multiple line tripping, breaker failures. The failure diagnosis algorithm is further enhanced to come up with the fully functional version of the failure diagnosis tool, which is tailored for the cases in which the PMU anomalies are present. In the developed algorithm the validity of the PMU data is critical; however, such causes as communication errors or cyber-attacks might lead to the PMU data anomalies. This issue is welladdressed in this paper and some major types of anomaly detection methods suitable for PMU data are discussed. Results show that the ensemble approach has some distinct advantages in data anomaly detection compared to the previously used standalone algorithms. Additionally, the enhanced failure diagnosis method is developed to clean the inaccurate data in case of the anomaly in measured voltage magnitudes. Finally, both original and enhanced versions of the tool are tested on 96-bus test system using the real-time OPAL-RT simulator. The results show the accuracy of the enhanced tool and its advantages over the primary version of the tool.

Data‐driven operation of the resilient electric grid: A case of COVID‐19

“…Finally, a model to find the root causes of the observed events assisted by the cyber log data from the protection devices are developed. Anomaly detection in PMUs is further discussed in [ 62 ] to detect the root cause of the failure in the transmission protection system, and it is shown that ensemble models are superior to stand‐alone methods.…”

“…Three infected IT security engineers at a nuclear power plant were quarantined for 14 days [19], increasing the risks that threaten normal operations. State of the art technologies in AI and ML can aid in reducing humans' direct involvement with overseeing, maintaining, and failure detection [57]. Protection systems are always of the highest priority in the energy industry.…”

Denoising and Bad Data Detection in Distribution Phasor Measurements using Filtering, Clustering and Koopman Mode Analysis

“…Data pre-processing and measurement denoising is a significant step which needs to be accomplished prior to investigation of any physical or cyber-induced events throughout the system [1]. A proper BDD technique is of enormous concern in such Distribution System (DS) applications as State Estimation (SE) [2], fault classification [3], fault location identification [4], and situational awareness [5]. [6] presents a comprehensive study on different ML-based approaches for distribution level disturbance analysis with regard to efficacy examination and performance evaluation of each technique.…”

Iterative Online Fault Identification Scheme for High-Voltage Circuit Breaker Utilizing a Lost Data Repair Technique

“…In [13], when processing the training set of SVM, a clustering method is used to identify the abnormal data, and the separated abnormal data are directly removed, which may destroy the continuity and integrity of the data. Reference [14] proposes two means to address the measurement anomalies: one is flagging them and the other is replacing the anomaly with the average value of the data points lying around. In [15], the abnormal data are imputed, utilizing the artificial neural network (ANN) and autoregressive exogenous input (ARX) models.…”

“…In state-of-the-art works, the main concern for missing data recovery is accuracy [1][2][3][4][5][6][7][8][9][10][11][12][13][14][15][16][17][18]. For realistic power grids, the number of electrical infrastructures reaches tens or hundreds of thousands, generating nearly ten or hundred million monitoring data every year.…”