Real-time Faulted Line Localization and PMU Placement in Power Systems through Convolutional Neural Networks

Wenting Li, Deepjyoti Deka, Michael Chertkov, Meng Wang

Abstract

Diverse fault types, fast re-closures, and complicated transient states after a fault event make real-time fault location in power grids challenging. Existing localization techniques in this area rely on simplistic assumptions, such as static loads, or require much higher sampling rates or total measurement availability. This paper proposes a faulted line localization method based on a Convolutional Neural Network (CNN) classifier using bus voltages. Unlike prior data-driven methods, the proposed classifier is based on features with physical interpretations that improve the robustness of the location performance. The accuracy of our CNN based localization tool is demonstrably superior to other machine learning classifiers in the literature. To further improve the location performance, a joint phasor measurement units (PMU) placement strategy is proposed and validated against other methods. A significant aspect of our methodology is that under very low observability (7% of buses), the algorithm is still able to localize the faulted line to a small neighborhood with high probability. The performance of our scheme is validated through simulations of faults of various types in the IEEE 39-bus and 68-bus power systems under varying uncertain conditions, system observability, and measurement quality.

1. Introduction

Rapid fault identification and isolation are critical for maintaining the stability of modern power grids. When a short circuit or other fault occurs, operators must pinpoint the affected line immediately to prevent cascading failures. However, the increasing complexity of modern grids and the high cost of monitoring equipment present significant hurdles. Installing high-speed sensors, such as Phasor Measurement Units (PMUs), at every bus is often economically unfeasible.

Traditional localization techniques frequently struggle with two extremes: they either require a fully observed system with sensors at every node or rely on extremely high sampling rates (up to 1MHz) that exceed standard PMU capabilities. Furthermore, many existing data-driven approaches treat the grid as a "black box," ignoring the underlying electrical physics. This lack of physical grounding makes such models fragile when faced with changing load conditions or measurement noise.

2. Methodology

This research introduces a two-pronged approach that combines physical insights with deep learning to achieve robust localization under limited observability.

2.1 Physically-Informed Feature Extraction

Rather than feeding raw voltage data directly into a neural network, the proposed method utilizes Substitution Theory to derive a physically interpretable feature vector ($\psi$). By multiplying the pre-fault admittance matrix ($Y^0$) by the change in bus voltages ($\Delta U$), a vector is created that represents unbalanced currents. In an ideal system, this vector is sparse, exhibiting non-zero values only at the buses directly connected to the fault. This transformation simplifies the learning task for the neural network and enhances robustness against grid noise.

2.2 CNN-Based Classification

The core of the localization tool is a 4-layer Convolutional Neural Network (CNN). CNNs are uniquely suited for power system applications because their convolutional layers naturally mimic the local connectivity and topology of the grid. This architecture allows the model to learn spatial correlations between neighboring buses effectively.

The proposed 4-layer CNN architecture for fault localization.

2.3 Optimal PMU Placement

To maximize the utility of a limited number of sensors, a greedy PMU placement algorithm was developed. This strategy selects bus locations that minimize the CNN’s training loss while prioritizing "high-degree" buses—hubs with numerous connections. This ensures the model receives the most informative data possible from the network.

3. Experimental Evaluation

The methodology was validated using two standard benchmarks: the IEEE 39-bus system (representing the New England grid) and the more complex IEEE 68-bus system (representing the interconnected New York/New England systems). Data was generated using the Power System Toolbox (PST), simulating four primary fault types: Three-Phase (TP), Line-to-Ground (LG), Double-Line-to-Ground (DLG), and Line-to-Line (LL). The simulations accounted for varying load conditions, fault impedances, and measurement noise.

4. Results and Discussion

4.1 Comparative Performance

The CNN-based approach demonstrated significant superiority over traditional machine learning baselines, including Multi-class Support Vector Machines (MSVM) and standard Neural Networks (NN). As shown in the performance metrics, the CNN maintains high Localization Accuracy Rates (LAR) even as the percentage of measured buses decreases.

Comparison of Localization Accuracy Rate (LAR) between CNN, MSVM, and 2-NN across different fault types and observability levels.

In the 68-bus system, the CNN achieved over 90% accuracy with only 15% PMU coverage. Remarkably, even at extremely low observability (7%), the algorithm consistently identifies the correct line within its top 3 or top 5 predictions, providing a manageable search area for repair crews.

4.2 Robustness and Efficiency

The model proved resilient to various real-world challenges:

  • Noise: Performance remained stable with Signal-to-Noise Ratios (SNR) as low as 60dB.
  • Synchronization: The system handled measurement delays of up to 40ms.
  • Real-time Processing: Once trained, the localization process takes approximately 4 milliseconds, enabling true real-time response.

5. Conclusion

By integrating physical grid properties into a CNN-based framework, this research provides a robust and efficient solution for real-time fault localization. The method's ability to maintain high accuracy under low observability and its resilience to noise make it a promising tool for modern utility operators. Future work will focus on validating these results with real-world utility data and extending the methodology to the more complex, highly branched environments of distribution grids.

Summary updated

Abstract

Diverse fault types, fast re-closures, and complicated transient states after a fault event make real-time fault location in power grids challenging. Existing localization techniques in this area rely on simplistic assumptions, such as static loads, or require much higher sampling rates or total measurement availability. This paper proposes a faulted line localization method based on a Convolutional Neural Network (CNN) classifier using bus voltages. Unlike prior data-driven methods, the proposed classifier is based on features with physical interpretations that improve the robustness of the location performance. The accuracy of our CNN based localization tool is demonstrably superior to other machine learning classifiers in the literature. To further improve the location performance, a joint phasor measurement units (PMU) placement strategy is proposed and validated against other methods. A significant aspect of our methodology is that under very low observability (7% of buses), the algorithm is still able to localize the faulted line to a small neighborhood with high probability. The performance of our scheme is validated through simulations of faults of various types in the IEEE 39-bus and 68-bus power systems under varying uncertain conditions, system observability, and measurement quality.
BibTeX
@article{li2019real,
  author = {Wenting Li and Deepjyoti Deka and Michael Chertkov and Meng Wang},
  title = {{Real-time Faulted Line Localization and PMU Placement in Power Systems through Convolutional Neural Networks}},
  journal = {IEEE Transactions on Power Systems},
  year = {2019},
  doi = {10.1109/tpwrs.2019.2917794},
}
Figure