Generative AI-Enhanced Real-Time Anomaly Detection in Integrated Energy Systems

The integration of Integrated Energy Systems (IES) with main power grids is vital for enhancing the efficiency, security, and resilience of modern energy systems. However, this increased interconnectivity among energy sources, smart grids, and digital monitoring systems also exposes IES to significant cyber threats. To mitigate these risks, we have developed a framework for real-time, AI-assisted monitoring and anomaly detection, leveraging generative AI models to monitor interconnected IES within the main grid. This paper presents a Generative Adversarial Networks (GAN)-based prediction and anomaly detection system, specifically leveraging the Wasserstein GAN with Gradient Penalty (WGAN-GP) with Long Short-Term Memory (LSTM), for secure monitoring of grid-connected IES. Our approach combines the generator’s predictive capabilities with the discriminator’s scoring mechanism to enhance anomaly detection and overall model accuracy. This allows the system to forecast the control system’s future response and detect anomalies before they manifest in real-time data. The effectiveness of this framework is assessed using an interconnected IES to the IEEE 118-bus test network. The pre-trained model is subjected to diverse attack scenarios, and experimental results consistently demonstrate its capability to identify the probability of anomalies within the IES efficiently.