Embedded, Real-Time, and Distributed Traveling Wave Fault Location Method Using Graph Convolutional Neural Networks

Abstract

This work proposes and develops an implementation of a fault location method to provide a fast and resilient protection scheme for power distribution systems. The method analyzes the transient dynamics of traveling waves (TWs) to generate features using the discrete wavelet transform (DWT), which are then used to train several graph convolutional network (GCN) models. Faults are simulated in the IEEE 34-node system, which is divided into three protection zones (PZs). The goal is to identify the PZ in which the fault occurs. The GCN models create a distributed protection scheme, as all nodes are able to retrieve a prediction. Given that message-passing between nodes occurs both during training and in the execution of the model, the resiliency of such schemes to communication losses was analyzed and demonstrated. One of the models, which only uses voltage measurements, was implemented on a Texas Instruments F28379D development board. The execution times were monitored to assess the speed of the protection scheme. It is shown that the proposed method can be executed in approximately a millisecond, which is comparable to existing TW protection in the transmission system. For experimental purposes, a DWT-based detection method is employed. A design of a setup to playback TWs using two development boards is also addressed.

Protection Analysis of a Traveling-Wave, Machine-Learning Protection Scheme for Distributions Systems With Variable Penetration of Solar PV

“…The employed ML algorithm is a reducedsize Random Forest (RF), called "micro RF". This scheme was previously introduced and experimentally validated in [10], achieving fault detection and location times of about 1.2 milliseconds. However, this work expands on the detection stage and the TW, ML relays placement criteria.…”

“…Other works show that ML models can achieve accuracies of more than 99% for fault location using a distributed approach with VOLUME 4, 2016 Graph Convolutional Networks (GCNs) [53]. However, the hardware implementation of GCNs requires communication among devices [54]. In the same system, local methods (without any communication) can achieve around 90% accuracy on fault location [55], [56] under several scenarios of renewable penetration.…”

“…In both cases, the detection logic is based on a 1level DWT decomposition. Furthermore, scaled-down fault location models, with or without communication, that can be implemented on low-cost micro-controllers have already been proposed in [10], [54], using GCNs or RF, respectively. In both cases, the execution times are slightly over 1 ms.…”

“…Both approaches use a 3-level DWT for feature extraction. Regarding the accuracy of such scale-down models, the RF models in [10] achieve an accuracy larger than 96% for fault location in the IEEE 34 node system.…”

“…From the power system protection qualities previously exposed at the beginning of this section, both the computational speed and discrimination capabilities of the TW detection method with DWT have already been demonstrated in [57], [58], as well as the accuracy and speed of a fault location model with scaled-down RFs in [10]. This work is based upon those references and elaborates a detailed analysis of the sensitivity, selectivity, false-trip probability, and unnecessary load loss metrics for a protection scheme composed of such fast, local, and distributed relays.…”