A Heterogeneous Graph-Based Multi-Task Learning for Fault Event Diagnosis in Smart Grid

Architecture of the proposed heterogenous MTL-GNN. The input features go through the common backbone GNN to generate graph embeddings. The embeddings generated by 128 nodes are flattened and concatenated together to convert to a one-dimensional vector .The concatenated feature vector is passed to 5 heads, three classification heads and two regression heads.

A heterogeneous multi-task learning graph neural network (MTL-GNN) is proposed which is capable of detecting, locating and classifying faults in addition to providing an estimate of the fault resistance and current. It is evaluated with IEEE-123 feeder system.

Contributions

Proposed a unified heterogeneous MTL-GNN architecture to perform 5 different tasks simultaneously for a fault event which are fault detection, fault localization, fault type classification, fault resistance estimation and fault current estimation
The proposed model performs well in the presence of measurement error, variable resistance and topology changes as common factors considered in real-world deployments
An explainability algorithm specific to GNN is utilized to identify key nodes in the grid which provides the opportunity for informed sparse measurement

IEEE-123 Feeder System

To evaluate the proposed method, we utilize the IEEE-123 feeder system which is shown below. Taking into account the voltage regulators the total number of nodes ends up being 128 instead of 123. More specific details about the feeder system are available here .

Diagram of IEEE-123 node feeder system. The highlighted blue blocks represent the node pairs that are considered connected. The number of voltage regulators, transformers, and switches is mentioned in (·) and the number of buses, their phases and load connectivity are mentioned in the table.

There are a total of 68 three-phase nodes which are considered for the dataset generation. Based on previous research works we make the assumption that some specific pairs of nodes are connected which are (149, 150r), (18, 135), (13, 152), (60, 160, 160r,), (61, 61s), (97, 197), (9, 9r), (25, 25r).

Dataset Generation

OpenDSS is used to generate dataset for 5 different fault types. This includes asymmetrical faults consisting of line-to-ground faults (LG), line-to-line faults (LL), line-to- line-to-ground faults (LLG) and symmetrical faults consisting of line-to-line-to-line-to-ground faults (LLLG), line-to-line- to-line faults (LLL). Both fault resistance and connected load values are sampled from two different uniform distribution.

Visualization of the sequence of procedures for a single data point and label generation. The double-circled digits represent the loop iteration points in the algorithm. By performing all the iterations in a hierarchical manner (the digits specify the order of the hierarchy) the entire dataset is generated

For each node in the IEEE-123 feeder system fault is simulated for a specific fault type and fault resistance. The generated data point contain 3-phase voltage magnitude and phase

Confidence level of Predictions

Another practical aspect to consider is how the grid operators will manage the system on a day-to-day basis. To ensure the reliability of model predictions, grid operators need access to more information.

Envisioned user interface for grid operators. (Left) Detected fault location (in this example node 77) in the distribution grid along with 1-hop nodes (in this example 76, 78) and 2-hop nodes (in this example 72, 86, 79, 80) in the neighborhood of the predicted node. (Right) The confidence level behind the predictions provides grid operators with additional information to verify the reliability of the prediction and fault resistance and fault current estimation allow grid operators to take appropriate safety measures for fault isolation and clearance.

The grid operators will be provided by both the predicted location and the neighboring nodes within 1-hop and 2-hop distances. In addition, grid operators will have access to the voltage phasors of these nodes which provides more context behind the predictions. Furthermore, the confidence level of these predictions will allow grid operators to make an informed decision and undertake necessary safety measures before taking any action

Extension of the Following Work

This work is extension of the following work that was published in MLSP 2023.

MLSP (Chanda et. al 2023) Chanda, Dibaloke, and Nasim Yahya Soltani. "Graph-Based Multi-Task Learning For Fault Detection In Smart Grid." In 2023 IEEE 33rd International Workshop on Machine Learning for Signal Processing (MLSP), pp. 1-6. IEEE, 2023.

Project Page Poster

Abstract: Timely detection of electrical faults is of paramount importance for efficient operation of the smart grid. To better equip the power grid operators to prevent grid-wide cascading failures, the detection of fault occurrence and its type must be accompanied by accurately locating the fault. In this work, we propose a multi-task learning architecture that encodes the graph structure of the distribution network through a shared graph neural network (GNN) to both classify and detect faults and their locations simultaneously. Deploying GNNs allows for representation learning of the grid structure which can later be used to optimize grid operation. The proposed model has been tested on the IEEE-123 distribution system. Numerical tests verify that the proposed algorithm outperforms existing approaches.

How To Cite This Work

Use the following information to cite the paper.

Bibliography

Chanda, Dibaloke, and Nasim Yahya Soltani. "A Heterogeneous Graph-Based Multi-Task Learning for Fault Event Diagnosis in Smart Grid." IEEE Transactions on Power Systems (2024).

Bibtex

@article{chanda2024heterogeneous,
                title={A Heterogeneous Graph-Based Multi-Task Learning for Fault Event Diagnosis in Smart Grid},
                author={Chanda, Dibaloke and Soltani, Nasim Yahya},
                journal={IEEE Transactions on Power Systems},
                year={2024},
                publisher={IEEE}
              }