-
Time-Synchronized State Estimation Using Graph Neural Networks in Presence of Topology Changes
Authors:
Shiva Moshtagh,
Anwarul Islam Sifat,
Behrouz Azimian,
Anamitra Pal
Abstract:
Recently, there has been a major emphasis on develo** data-driven approaches involving machine learning (ML) for high-speed static state estimation (SE) in power systems. The emphasis stems from the ability of ML to overcome difficulties associated with model-based approaches, such as handling of non-Gaussian measurement noise. However, topology changes pose a stiff challenge for performing ML-b…
▽ More
Recently, there has been a major emphasis on develo** data-driven approaches involving machine learning (ML) for high-speed static state estimation (SE) in power systems. The emphasis stems from the ability of ML to overcome difficulties associated with model-based approaches, such as handling of non-Gaussian measurement noise. However, topology changes pose a stiff challenge for performing ML-based SE because the training and test environments become different when such changes occur. This paper circumvents this challenge by formulating a graph neural network (GNN)-based time-synchronized state estimator that considers the physical connections of the power system during the training itself. The results obtained using the IEEE 118-bus system indicate that the GNN-based state estimator outperforms both the model-based linear state estimator and a data-driven deep neural network-based state estimator in the presence of non-Gaussian measurement noise and topology changes, respectively.
△ Less
Submitted 31 May, 2023; v1 submitted 8 December, 2022;
originally announced December 2022.
-
Data-Driven Flow and Injection Estimation in PMU-Unobservable Transmission Systems
Authors:
Satyaprajna Sahoo,
Anwarul Islam Sifat,
Anamitra Pal
Abstract:
Fast and accurate knowledge of power flows and power injections is needed for a variety of applications in the electric grid. Phasor measurement units (PMUs) can be used to directly compute them at high speeds; however, a large number of PMUs will be needed for computing all the flows and injections. Similarly, if they are calculated from the outputs of a linear state estimator, then their accurac…
▽ More
Fast and accurate knowledge of power flows and power injections is needed for a variety of applications in the electric grid. Phasor measurement units (PMUs) can be used to directly compute them at high speeds; however, a large number of PMUs will be needed for computing all the flows and injections. Similarly, if they are calculated from the outputs of a linear state estimator, then their accuracy will deteriorate due to the quadratic relationship between voltage and power. This paper employs machine learning to perform fast and accurate flow and injection estimation in power systems that are sparsely observed by PMUs. We train a deep neural network (DNN) to learn the map** function between PMU measurements and power flows/injections. The relation between power flows and injections is incorporated into the DNN by adding a linear constraint to its loss function. The results obtained using the IEEE 118-bus system indicate that the proposed approach performs more accurate flow/injection estimation in severely unobservable power systems compared to other data-driven methods.
△ Less
Submitted 8 December, 2022;
originally announced December 2022.
-
Monitoring and Detection of Low-current High-Impedance Faults in Distribution Networks
Authors:
Anwarul Islam Sifat,
Fiona J. Stevens McFadden,
Ramesh Rayudu,
Joseph Bailey
Abstract:
Faults in electricity distribution networks have the potential to ignite fires, cause electrocution, and damage the system itself. High current Low Impedance Faults (LIF) are typically detected and mitigated via over-current, distance, directional relays, fuses, etc. In contrast, while High Impedance Faults (HIF) are equally hazardous, they are much more challenging to detect due to the fault curr…
▽ More
Faults in electricity distribution networks have the potential to ignite fires, cause electrocution, and damage the system itself. High current Low Impedance Faults (LIF) are typically detected and mitigated via over-current, distance, directional relays, fuses, etc. In contrast, while High Impedance Faults (HIF) are equally hazardous, they are much more challenging to detect due to the fault current being much lower than load currents and their time-varying and nonlinear behaviour. Moreover, New Zealand distribution networks are extensive and largely unmonitored beyond the substation, and suitable HIF detection schemes are still an ongoing research challenge.
To date, we have built a physical test facility for power system fault analysis and develo** and evaluating our sensing and fault detection system. We have simulated LIF and HIF with different fault surface materials and load-switching events. From the data collected, we have characterized the unique fault behaviour for both LIF and HIF in 400V networks and trained a Deep Learning classifier to recognize the type of fault present from its unique signature. We have developed an outdoor pole mountable sensing system and have installed this in Wellington Electricity's network for ongoing data collection and evaluation.
This paper will describe the test facility and our experience develo** and implementing the sensing system. The widest range of HIF phenomena observed was in the fault experiments involving the tree branch. For brevity, therefore, this paper reports on the results of just these tree-branch experiments. HIF faults on other surface materials will be reported elsewhere. Finally, we will detail the pole-mountable sensing system installed in Wellington Electricity's network and the outcomes thus far.
△ Less
Submitted 30 October, 2022;
originally announced October 2022.
