A Hybrid CNN-LSTM Approach Dynamic Security Assessment of Power Systems with GAN-based Imbalanced Database

Azad, Sasan; Ameli, Mohammad Taghi; Ameli, Hossein; Strbac, Goran

doi:10.48308/ijrtei.2024.237543.1062

A Hybrid CNN-LSTM Approach Dynamic Security Assessment of Power Systems with GAN-based Imbalanced Database

Document Type : Research Article

Authors

Sasan Azad ¹

Mohammad Taghi Ameli ¹

Hossein Ameli ²

Goran Strbac ²

¹ Department of Electrical Engineering, Abbaspour School of Engineering, Shahid Beheshti University, Tehran, Iran

² Control and Power Group, Imperial College London, London SW7 2AZ, UK

10.48308/ijrtei.2024.237543.1062

Abstract

Recently, deep learning-based techniques in dynamic security assessment (DSA) have shown significant advances, enabling them to play a pivotal role in ensuring power systems' secure operation. However, imbalanced samples are a fundamental challenge for effective training of data-driven methods. In the DSA problem, especially in real-world power systems, the database is usually imbalanced and the number of secure cases is more than the number of insecure cases. This imbalance can lead to loss of fit and generalization in insecure cases since the DSA model tends to focus too much on secure cases. In past studies, methods based on linear interpolators have been used, which cannot satisfy the power system's physical characteristics. This paper addresses the data imbalance in DSA by using generative adversarial networks (GANs) to generate synthetic data resembling the original data. After addressing the data imbalance, a hybrid model consisting of a convolutional neural network (CNN) and long short-term memory (LSTM) is developed in an integrated framework for DSA. The proposed model was implemented and tested on the IEEE 39 bus system. The test results show that solving the data imbalance problem has improved the proposed DSA model's performance.

Keywords

Dynamic security assessment

Deep learning

Generative adversarial network

Convolutional neural network

Long short-term memory

Imbalanced data

Subjects