SynEnergy Generates Anomaly-Preserving Synthetic Energy Data

Access to fine-grained energy consumption data is crucial for various applications, including demand forecasting, response planning, and grid reliability. However, privacy concerns and data-sharing restrictions often limit the availability of such data, driving interest in synthetic data generation. While existing methods can reproduce overall consumption patterns, they frequently smooth out or underrepresent critical anomalous events caused by factors like extreme weather or infrastructure failures. These anomalies are challenging to preserve due to their sparsity, localized nature, and complex dependencies. To address this, researchers propose SynEnergy, a two-stage diffusion-based framework designed for anomaly-preserving energy consumption data generation. The first stage, Heterogeneous Graph-based Anomaly Semantic Learning (HG-ASL), extracts region-specific anomaly semantics. It achieves this by jointly modeling spatial and attribute dependencies across urban regions from sparse residual structures. The second stage, Anomaly Semantic-guided Diffusion (AS-Diff), then injects these learned anomaly semantics directly into the denoising process of a diffusion model. This allows for the generation of realistic consumption sequences that accurately preserve anomalous patterns. The framework supports controllable generation for individual regions and scales efficiently to city-wide settings. Evaluations on four real-world energy datasets show SynEnergy significantly improves anomaly preservation fidelity by an average of 12.21% and downstream quality by 2.96%, while maintaining competitive overall generation fidelity compared to 11 baselines.