[Paper Review] Data-Driven Short-Term Voltage Stability Assessment Based on Spatial-Temporal Graph Convolutional Network

Yonghong Luo, Chao Lü+2 more|arXiv (Cornell University)|Mar 5, 2021

Optimal Power Flow Distribution33 references4 citations

TL;DR

This paper proposes a spatial-temporal graph convolutional network (STGCN) for data-driven short-term voltage stability (SVS) assessment, leveraging graph convolution to model electrical topology and 1D convolution to capture temporal dynamics. The STGCN achieves higher accuracy, robustness to noise, and adaptability to topology changes and operating point variations than existing methods, with 98.7% testing accuracy on the Guangdong Power Grid.

ABSTRACT

Post-fault dynamics of short-term voltage stability (SVS) present spatial-temporal characteristics, but the existing data-driven methods for online SVS assessment fail to incorporate such characteristics into their models effectively. Confronted with this dilemma, this paper develops a novel spatial-temporal graph convolutional network (STGCN) to address this problem. The proposed STGCN utilizes graph convolution to integrate network topology information into the learning model to exploit spatial information. Then, it adopts one-dimensional convolution to exploit temporal information. In this way, it models the spatial-temporal characteristics of SVS with complete convolutional structures. After that, a node layer and a system layer are strategically designed in the STGCN for SVS assessment. The proposed STGCN incorporates the characteristics of SVS into the data-driven classification model. It can result in higher assessment accuracy, better robustness and adaptability than conventional methods. Besides, parameters in the system layer can provide valuable information about the influences of individual buses on SVS. Test results on the real-world Guangdong Power Grid in South China verify the effectiveness of the proposed network.

Motivation & Objective

• To address the limitation of existing data-driven methods in capturing the spatial-temporal dynamics of short-term voltage instability (SVS).
• To develop a deep learning model that effectively integrates electrical network topology and time-series PMU data for real-time SVS assessment.
• To improve model robustness against noise, topology changes, and varying operating conditions.
• To provide interpretable insights into the influence of individual buses on system stability through system-layer parameters.

Proposed method

• Uses Chebyshev graph convolution to embed network topology into the model, capturing spatial locality in voltage instability propagation.
• Applies one-dimensional convolutional layers to extract temporal features from post-fault PMU time series.
• Employs spatial-temporal feature fusion blocks to combine multi-receptive field representations across space and time.
• Introduces a two-stage architecture: a node layer for bus-level feature learning and a system layer for global stability classification.
• Leverages the graph Laplacian and Chebyshev polynomial approximation to efficiently compute graph convolutions.
• Trains the end-to-end STGCN model using labeled PMU data from fault scenarios to classify system stability.

Experimental results

Research questions

• RQ1 Can a deep learning model effectively capture the spatial-temporal dynamics of short-term voltage instability using PMU data?
• RQ2 How does incorporating electrical network topology improve the accuracy and robustness of data-driven SVS assessment compared to geographical or no topology modeling?
• RQ3 To what extent does the proposed STGCN model maintain performance under noisy measurements, topology changes, and varying operating points?
• RQ4 Can the system-layer parameters of the STGCN model provide interpretable insights into the contribution of individual buses to voltage stability?

Key findings

• The proposed STGCN achieved 98.7% testing accuracy under different operating points, outperforming ST-shapelet (97.3%), LSTM (92.0%), and RVFL (92.3%) methods.
• In noisy environments, the STGCN maintained 98.6% accuracy, significantly higher than LSTM (92.0%) and RVFL (84.0%), demonstrating superior noise robustness.
• Under topology changes, the STGCN achieved 96.0% accuracy, surpassing ST-shapelet (93.0%), LSTM (94.0%), and RVFL (93.0%) methods due to its topology-aware design.
• The model’s performance remains stable across diverse fault scenarios, indicating strong adaptability to system variations.
• The system-layer parameters provide interpretable information on the influence of individual buses on overall voltage stability, enabling diagnostic insights.
• The integration of Chebyshev graph convolution with 1D temporal convolutions enables effective modeling of both spatial locality and temporal evolution in SVS.