Dynamic Security Assessment under Topology Changes and Missing Data Using a Multi-Task Transfer Learning Framework with Focus on Intelligent Load Shedding

Poormoradi, Nazanin; Ameli, Mohammad Taghi

doi:10.48308/ijrtei.2025.240270.1090

Dynamic Security Assessment under Topology Changes and Missing Data Using a Multi-Task Transfer Learning Framework with Focus on Intelligent Load Shedding

Document Type : Research Article

Authors

Nazanin Poormoradi

Mohammad Taghi Ameli

Department of Electrical Engineering, Abbaspour School of Engineering, Shahid Beheshti University, Tehran, Iran

10.48308/ijrtei.2025.240270.1090

Abstract

Data-driven methods for DSA, particularly those based on deep learning (DL), have shown promising results. However, these approaches face challenges such as topology modifications and missing data, which hinder their implementation in real-world power systems. On the other hand, most existing DSA methods function as black-box models, offering limited interpretability and failing to provide actionable insights or recommended corrective measures to prevent power system instability. This paper proposes a multitask transfer learning framework to address these challenges. In this framework, three separate models are trained using a multitask learning approach. Each model is responsible for a specific task, including missing data reconstruction, DSA, and intelligent load shedding (LS). Generative adversarial networks, an unsupervised DL technique based on the competition between two neural networks, are employed to handle missing data. For DSA and LS, graph convolutional networks are employed to enhance accuracy by effectively capturing the topological structure of the power system. The multitask learning framework for data reconstruction, DSA, and LS is well-trained for a specific topology; however, it experiences a decrease in performance accuracy when applied to unseen topologies. To address this challenge, transfer learning is utilized to update the model with a small dataset. Implementing the proposed framework on the IEEE 39-bus system shows its positive performance in solving the challenge of missing data and topology changes.

Keywords

Dynamic security assessment

Topology change

Graph convolutional network

Transfer learning

Multitask learning

Missing data

Subjects