Missing Data Imputation for Carbon Emission Factor Calculation in Power Systems with Enhanced Generative Adversarial Networks

To address the challenges of missing measurement data in modern power systems, which critically hinders the accurate calculation of carbon emission factors, this paper proposes a physics-constrained Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP) for dynamic data imputation. Conventional approaches, including ARIMA, KNN, and standard GANs, exhibit limitations in modeling complex missing-data scenarios and maintaining physical consistency with power system dynamics. Our enhanced framework establishes a generator-discriminator adversarial architecture that synergistically integrates a physics-informed regularization term with the stabilized training mechanism of WGAN-GP. The innovation lies in the dual-constrained optimization: (1) gradient penalty enforcement through the discriminator to ensure Lipschitz continuity, and (2) physical constraint embedding via customized loss functions that preserve fundamental electrical conservation laws. Implemented on the IEEE 39-bus system, the proposed method demonstrates superior performance under 10–50% random missing rates, achieving 23.7% lower MAE and 34.2% reduced RMSE compared to BiLSTM. This advancement provides a novel paradigm for intelligent recovery of heterogeneous measurement data essential for reliable carbon emission assessment in cyber-physical power systems.