Concetti Chiave
A novel conditional diffusion model is proposed to generate high-quality, long-term synthetic building energy data by effectively incorporating relevant metadata such as meter types and building types.
Sintesi
The study introduces a comprehensive framework that integrates meta-information into conditional generative models to generate synthetic building energy data. This reduces dependency on abundant historical data and eliminates the need for laborious parameter tuning, which is a challenge commonly faced with traditional methods such as regression models and building performance simulation (BPS).
The key highlights and insights are:
The study proposes a conditional diffusion model for generating high-quality synthetic energy data using relevant metadata such as location, weather, building, and meter type. This model is compared with traditional methods like Conditional Generative Adversarial Networks (CGAN) and Conditional Variational Auto-Encoders (CVAE).
The conditional diffusion model explicitly handles long-term annual consumption profiles, producing coherent synthetic data that closely resembles real-world energy consumption patterns. The results demonstrate the proposed diffusion model's superior performance, with a 36% reduction in Fréchet Inception Distance (FID) score and a 13% decrease in Kullback-Leibler divergence (KL divergence) compared to the following best method.
The proposed method successfully generates high-quality energy data through metadata, and its code will be open-sourced, establishing a foundation for a broader array of energy data generation models in the future.
The study evaluates the models using the extensive Building Data Genome 2.0 (BDG2) dataset, which comprises power meters from around the world, enabling the assessment of the models' ability to adapt to the wide-ranging characteristics of energy data.
The evaluation metrics encompass FID, KL divergence, Root Mean Squared Error (RMSE), and Coefficient of Determination (R2) to provide a comprehensive analysis of the generative models' performance in terms of diversity, distribution similarity, and time-series prediction accuracy.
Statistiche
The proposed conditional diffusion model achieved a FID score of 517.3 ± 2.1, a KL divergence of 0.40 ± 0.0013, an RMSE of 0.25 ± 0.00052, and an R2 of 0.43 ± 0.0021.
Citazioni
"The incorporation of conditional variables like building type and meter type distinguishes CVAEs and guides the data generation process."
"CGANs are adept at capturing complex data distributions by utilizing adversarial loss instead of relying on restrictive probabilistic models."
"Diffusion models offer advantages in training stability, ease of hyperparameter tuning, and high sample quality compared to traditional generative approaches like GANs and VAEs."