Prediction of Node Importance of Power System Based on ConvLSTM

Xu Wu, Junqi Geng, Meng Liu, Zongxun Song, Huihui SongEnergies 2022 · DOI

TL;DR

This paper addresses the critical need for predicting power system node importance to prevent cascading failures. It introduces a ConvLSTM-based model that leverages both spatial and temporal features of power grids, outperforming traditional LSTM in accuracy and providing crucial early warnings for grid stability.

What Problem Does This Paper Address?

Modern power systems are increasingly complex and interconnected, making them vulnerable to cascading failures initiated by the failure of critical nodes. Current methods for identifying important nodes rely on real-time power flow calculations, which inherently lag behind the dynamic state of the grid. This delay prevents timely protection of vulnerable nodes and hinders the ability to predict and prevent blackouts. The paper aims to bridge this gap by developing a predictive model for node importance that can provide early warnings and enable proactive risk mitigation.

Methods

The core of this research is the development of a ConvLSTM (Convolutional Long Short-Term Memory) prediction model. Unlike standard LSTM networks that process sequential data using fully connected layers, ConvLSTM incorporates convolutional operations. This allows the model to simultaneously capture:

  • Temporal dependencies: Similar to LSTM, it can learn from historical data sequences.
  • Spatial dependencies: The convolutional layers extract features from the grid's topology and the relationships between interconnected nodes.

This spatiotemporal feature extraction is crucial for understanding how node importance evolves and how changes in one part of the grid affect others. The authors also utilize the EB-SALSA algorithm to generate the ground truth data for node importance, which considers both authority (in-degree) and hub (out-degree) values within the power system's network topology.

Datasets

The study uses simulated datasets derived from two standard power system benchmarks:

  • IEEE-39 bus system: A smaller-scale system used for initial model development and validation.
  • IEEE-118 bus system: A larger-scale system to demonstrate the model's scalability and performance on more complex grids.

The node importance data for these systems is generated by repeatedly applying the EB-SALSA algorithm to power flow data.

Key Insights and Ideas

  • Spatiotemporal prediction is key: Power system node importance is not just a function of time but also of its spatial relationships within the grid. ConvLSTM's ability to capture both is a significant advantage.
  • Lagging indicators are insufficient: Relying on real-time power flow for node importance assessment is inherently reactive. Predictive models are essential for proactive grid management.
  • ConvLSTM's architectural advantage: Replacing LSTM's fully connected layers with convolutional layers allows for efficient extraction of spatial features, leading to more accurate predictions.
  • Scalability matters: The proposed ConvLSTM model demonstrates effectiveness across different power system scales, suggesting its potential for widespread application.

Key Results

The ConvLSTM model consistently outperformed the standard LSTM model in predicting node importance across both the IEEE-39 and IEEE-118 bus systems. This was quantitatively demonstrated through several performance evaluation metrics:

  • IEEE-39 Bus System:
    • MAE (Mean Absolute Error): ConvLSTM (0.0222) vs. LSTM (0.0464)
    • RMSE (Root Mean Square Error): ConvLSTM (0.0298) vs. LSTM (0.0475)
    • MAPE (Mean Absolute Percentage Error): ConvLSTM (0.3081) vs. LSTM (1.0433)
  • IEEE-118 Bus System:
    • MAE: ConvLSTM (0.0133) vs. LSTM (0.0693)
    • RMSE: ConvLSTM (0.0150) vs. LSTM (0.0768)
    • MAPE: ConvLSTM (0.2274) vs. LSTM (1.1901)

These results indicate that ConvLSTM provides significantly more accurate predictions with lower error rates.

Limitations and Future Work

  • Data Generation: The study relies on simulated data generated by the EB-SALSA algorithm. While this provides a controlled environment, real-world power system data might present additional complexities.
  • Model Complexity: While ConvLSTM offers advantages, it can be computationally more intensive than simpler models.
  • Future Work: The authors suggest that the developed ConvLSTM model has universality and can be applied to power systems of different scales. Further research could explore its application to real-time grid operations, integration with other predictive maintenance strategies, and the impact of different types of disturbances on node importance.

Summary updated

Abstract

In power systems, the destruction of some important nodes may cause cascading faults. If the most important node in the power system can be found, the important node can be protected in advance, thereby avoiding a blackout accident. At present, the evaluation algorithm of node importance is calculated based on the power flow of the power grid, so the calculation results must be lagging behind, and it does not have the ability to provide early warning for the power grid to provide protection signals. Therefore, it is necessary to predict the importance of nodes in the power system. After using a reasonable prediction model to predict the importance of nodes, we can simulate the future state of power system operation and avoid accidents for the dispatching agency of the power grid company according to the prediction results. This paper proposes a prediction model based on convolutional long short-term memory (ConvLSTM) to predict the importance of nodes. This method has obvious advantages over the long short-term memory (LSTM) network. The convolution operation is used to replace the original full connection operation of the LSTM network, which not only utilizes the advantages of convolution to extract spatial features but also retains the ability of LSTM to process time-series features. The simulation results show that the prediction of node importance using the ConvLSTM network is much more accurate than LSTM. Using statistical indicators to compare and analyze the prediction results, it can be seen that ConvLSTM has higher prediction accuracy. Therefore, using the ConvLSTM model to predict node importance has certain significance for grid dispatching agencies to accurately simulate the future state of the power system and avoid risks in advance.
BibTeX
@inproceedings{wu2022prediction,
  author = {Xu Wu and Junqi Geng and Meng Liu and Zongxun Song and Huihui Song},
  title = {{Prediction of Node Importance of Power System Based on ConvLSTM}},
  booktitle = {Energies},
  year = {2022},
  doi = {10.3390/en15103678},
}
Figure