Original scientific paper

https://doi.org/10.17559/TV-20241026002094

Integrating Meteorological Data and Bidirectional LSTM Models for Vulnerability Detection and Security Strategy Development in Power Energy Systems

Hongxia Wang ; State Grid Xinjiang Company Limited Electric Power Research Institute, Urumqi 830000, Xinjiang, China; Xinjiang Key Laboratory of Extreme Environment Operation and Testing Technology for Power Transmission & Transformation Equipment, Urumqi 830000, Xinjiang, China
Jie Xu ; Xinjiang Meteorological Service Center, Urumqi 830002, Xinjiang, China *
Yang Yang ; State Grid Xinjiang Company Limited Electric Power Research Institute, Urumqi 830000, Xinjiang, China; Xinjiang Key Laboratory of Extreme Environment Operation and Testing Technology for Power Transmission & Transformation Equipment, Urumqi 830000, Xinjiang, China
Meng Li ; State Grid Xinjiang Company Limited Electric Power Research Institute, Urumqi 830000, Xinjiang, China; Xinjiang Key Laboratory of Extreme Environment Operation and Testing Technology for Power Transmission & Transformation Equipment, Urumqi 830000, Xinjiang, China

* Corresponding author.

Abstract

The increasingly complex power energy system, coupled with the increasing frequency of extreme weather events, requires powerful models for vulnerability detection and security strategy development. This study proposes a novel approach that combines meteorological data with a bidirectional Long Short Term Memory (LSTM) model enhanced by attention mechanisms. This model effectively captures the dynamic interaction between time dependence, meteorological factors, and system vulnerabilities. The key performance indicators, including accuracy, precision, and recall, demonstrate that the model has excellent predictive ability compared to traditional methods. The results show that the average accuracy of the bidirectional LSTM model for vulnerability classification of power energy systems is as high as 97.3%, and the recall fluctuation in different testing environments is only 0.2%, which is superior to traditional techniques such as Unidirectional LSTM and Support Vector Machine (SVM). Numerical experiments have verified the robustness of the model in different scenarios, while case studies have demonstrated its practical applicability in identifying key vulnerabilities and notifying targeted security measures. The unique combination of meteorological data and power system operation data enables the bidirectional LSTM model to accurately analyze complex environmental impacts and effectively identify potential vulnerabilities. This study contributes to the development of energy system prediction and analysis, providing a scalable and adaptable framework to enhance the system's ability to resist meteorological impacts. Future work will focus on integrating real-time monitoring systems and extending the applicability of models to a wider range of energy infrastructure environments.

Keywords

attention mechanism; bidirectional long short-term memory; meteorological dana; power energy system; security protection strategies; vulnerability analysis

Hrčak ID:

337715

URI

https://hrcak.srce.hr/337715

Publication date:

31.10.2025.

Visits: 106 *