This study proposes a Generative Adversarial Network (GAN)-based method to detect abnormal energy deception attacks in CPS. By leveraging the adversarial training paradigm, our GAN-based approach learns to generate realistic energy consumption data while distinguishing between genuine and deceptive patterns. The proposed approach can capture complex temporal dependencies in energy consumption data and robustness against adversarial attacks. Experimental results on real-world datasets confirm the efficacy of detecting anomalous energy deception attacks, highlighting its potential for improving the security and resilience of critical energy infrastructure. Our approach is generic one that can be applied to other CPS to detect false data injection-based threats.