Voltage stability analysis in IBR-dominated power grids using an explainable AI-based optimized deep learning framework

Modern power systems are undergoing a significant transformation, moving away from traditional setups dominated by synchronous generators toward configurations with a high penetration of inverter-based resources (IBRs) including both grid-forming and grid-following inverters. This evolution presents unique challenges, particularly in the realm of voltage stability, as conventional indices and methods often fall short in capturing the complex dynamics of IBR-dominated networks. To address these limitations, this paper introduces a novel voltage stability assessment index specifically designed for power systems with substantial IBR penetration, which incorporates the distinctive reactive power and apparent power constraints of IBRs, providing a more accurate representation of stability margins. Building upon this foundation, an optimized deep learning framework is proposed, utilizing a convolutional neural network (CNN)-bidirectional long short-term memory (BDLSTM) network enhanced with an attention mechanism to significantly improve prediction accuracy. The hyperparameters of the CNN-BDLSTM model are meticulously fine-tuned using the arithmetic optimization algorithm to achieve an ideal balance between accuracy and computational efficiency. Furthermore, explainable artificial intelligence is integrated to evaluate and validate the influence of input features, enhancing the interpretability and transparency of the proposed approach. The effectiveness of the methodology is demonstrated through validation on IEEE 39-bus, IEEE 57-bus, and IEEE 145-bus systems. The results underscore the superior accuracy and computational efficiency of the proposed method, affirming its suitability for real-time voltage stability assessment in modern power systems characterized by high IBR penetration.