SocioDiff: A Socio-Aware Diffusion Model for Residential Electricity Consumption Data Generation

Disadvantaged communities face disproportionate energy burdens, yet they remain underrepresented in electricity consumption datasets. This lack of data may limit the development of targeted assistance programs and energy policymaking, thereby perpetuating systemic disparities and increasing the risk of energy poverty. To address this pressing issue, we propose SocioDiff, a novel diffusion-based model for generating synthetic residential electricity consumption data that incorporates socio-demographic information. SocioDiff introduces the Socio-aware SSSD process, which integrates social attributes into the data generation workflow. SocioDiff employs a two-step diffusion process to separately capture global trends and household-specific details to ensure representative and diverse synthetic consumption data. Additionally, adversarial training is used to improve the representation of disadvantaged communities. Experimental results using the Irish Energy Regulatory Commission dataset show that SocioDiff achieves a 63.59% improvement in Maximum Mean Discrepancy compared to traditional methods and leads to a 4.4% improvement in downstream load forecasting performance. These findings demonstrate the effectiveness of incorporating socio-demographic features for generating more equitable and inclusive consumption data and supporting the development of fairer energy models and policies for disadvantaged communities. The generated dataset and documentation are publicly available at https://github.com/Intelligame/SocialDiff.