Integrated attentive Bi-LSTM conditional GAN for power system oscillation localization

Abstract

This paper presents an advanced deep learning framework that combines the Transformer model with either Long Short-Term Memory (LSTM) or Bidirectional Long Short-Term Memory (Bi-LSTM) in a Conditional Generative Adversarial Network (cGAN) architecture. This innovative framework is specifically designed to address forced oscillation (FO) source localization in power systems. The proposed methodology makes two key contributions to the field. Firstly, synthetic time series data during FO occurrences is generated by integrating the Transformer architecture with LSTM/ Bi-LSTM neural networks. Secondly, the cGAN’s Discriminator-Classifier component is employed to predict the location of the oscillation source. The experimental results validate our framework’s performance in two areas: generating synthetic oscillation datasets and enhancing FO source identification accuracy compared to traditional approaches. The proposed Transformer Bi-LSTM cGAN architecture outperforms the existing methods, particularly in challenging situations with deliberate faults and mixed fault scenarios, achieving an 87% to 96% accuracy in oscillation source identification, thus validating its viability for real-world power grid deployment.