Diffusion model generates realistic yearly household electricity data
LoaDiff: Conditional Generation of Electricity Consumption Time Series for Energy Analytics
Machine LearningArtificial Intelligence
Summary
Household electricity use is changing, but real smart-meter data is hard to get due to privacy rules. The authors created LoaDiff, a computer model that can produce realistic fake electricity use data for a whole year measured every few minutes. This model can include details like what appliances a home has and weather conditions. They tested LoaDiff on several datasets and found it creates diverse and believable data that can help with energy analysis tasks like predicting electricity use or identifying appliances.
What this means in practice
- •For energy utilities: Generate realistic synthetic electricity consumption data to improve load forecasting models without using sensitive customer information.
- •For energy analytics teams: Create diverse electricity usage profiles conditioned on household and environmental factors for demand response and appliance detection research.
Authors
Mariia Baranova, Adrien Petralia, Etienne Le Naour, Nathan Etourneau, Guillaume Hofmann, Themis Palpanas
Abstract
The energy transition is reshaping residential electricity consumption through the increasing adoption of distributed generation, electrified appliances, and demand-response programs. Understanding these evolving behaviors requires access to granular smart-meter data for applications such as load forecasting, appliance detection, and demand-side flexibility analysis. However, such data are subject to strict access restrictions and data-protection regulations. Thus, realistic synthetic alternatives are necessary. In this paper, we introduce LoaDiff, a diffusion-based generative model for year-long, sub-hourly smart-meter load curves. LoaDiff supports flexible conditioning on static household attributes, such as appliance ownership, and dynamic contextual variables, including calendar information and outdoor temperature. We evaluate the model against multiple generative baselines on three residential electricity-consumption datasets. Our experiments assess four complementary dimensions: fidelity and diversity, training-record memorization risk, downstream utility for load forecasting and appliance detection, and conditional controllability under alternative temperature conditions. The results show that LoaDiff generates realistic and diverse load profiles, achieves a favorable trade-off between generation quality and limited evidence of memorization, preserves information useful for downstream energy applications, and responds coherently to changes in conditioning variables.