TriHead-GAN generates realistic carbon emission time series with triple-head discriminator

TriHead-GAN addresses the critical scarcity of high-frequency city-level carbon emission data, which hinders deep learning models for climate policy and regulatory mechanisms like the EU Carbon Border Adjustment Mechanism. Existing GANs and diffusion models often fail to preserve cross-variable correlations between CO2 and co-emitted pollutants or meteorological factors, and produce overly smooth sequences lacking realistic step-wise variability. The triple-head discriminator uses a Wasserstein critic for distributional authenticity, leakage-free regression for cross-variable dependency, and adjacent-difference prediction for temporal smoothness. The generator combines global self-attention with local temporal convolution, per-step noise injection, and an anti-smoothing loss that matches first-difference statistics.

Experiments on the self-collected Changsha Carbon dataset, two public carbon datasets (China, US), and the ETTh1 benchmark show that TriHead-GAN achieves favorable performance over mainstream baselines on the vast majority of settings. Importantly, the generated synthetic windows improve downstream forecasting accuracy in low-resource carbon monitoring scenarios. This work provides a practical solution for generating realistic carbon emission time series, enabling better-informed climate policy and carbon trading mechanisms even when real data are sparse.