Robust uncertainty-aware attack-fault diagnosis in hydrogen-based energy system based on a copula-meta-learning approach

Active Distribution Power Systems (ADPS), including hydrogen-based renewable energy units, are becoming essential components of modern electrical grids. These systems integrate distributed energy resources with advanced communication technology to enhance efficiency, flexibility, and reliability. However, a significant challenge they face is distinguishing between malfunctions and cyberattacks, particularly False Data Injection (FDI) attacks, which can severely impact grid performance. Traditional fault detection methods often struggle to differentiate between these two causes, leading to inaccurate diagnoses and delayed responses, which jeopardize system reliability. This paper presents the GAN–LSTM method, which utilizes a hybrid model combining Long Short-Term Memory (LSTM) networks with Generative Adversarial Networks (GANs). This approach enhances the system’s ability to adaptively detect faults and attacks by learning from past incidents. A major obstacle the method encounters is the inherent uncertainty in data monitoring. ADPS data is often noisy, incomplete, or inconsistent, which can undermine model performance. Factors such as incomplete sensor readings, varying environmental conditions, and unstable network states contribute to this uncertainty. To address this issue, we employ the Copula method, which captures and models relationships within uncertain data. By using the Copula approach, we ensure that our model remains reliable and accurate, even when dealing with ambiguous or insufficient data. Experimental results indicate that the proposed GAN-LSTM-based GAN–LSTM model outperforms traditional techniques, achieving approximately 6–15% higher detection accuracy in differentiating between faults and cyberattacks. Furthermore, when compared to conventional machine learning techniques that do not account for uncertainty, our model’s detection accuracy improves in scenarios characterized by significant data uncertainty, such as missing or corrupted sensor data. These findings validate the effectiveness of the proposed approach, demonstrating its capability to provide precise and trustworthy diagnostics, even in uncertain situations. The method represents a promising strategy for enhancing the security and resilience of ADPS while ensuring efficient fault diagnosis and cyberattack detection.