CENTS: Generating synthetic electricity consumption time series for rare and unseen scenarios

TL;DR

The results highlight the efficacy of the proposed method in generating realistic household-level electricity consumption data, paving the way for training larger foundation models in the energy domain on synthetic as well as real-world data.

Abstract

Recent breakthroughs in large-scale generative modeling have demonstrated the potential of foundation models in domains such as natural language, computer vision, and protein structure prediction. However, their application in the energy and smart grid sector remains limited due to the scarcity and heterogeneity of high-quality data. In this work, we propose a method for creating high-fidelity electricity consumption time series data for rare and unseen context variables (e.g. location, building type, photovoltaics). Our approach, Context Encoding and Normalizing Time Series Generation, or CENTS, includes three key innovations: (i) A context normalization approach that enables inverse transformation for time series context variables unseen during training, (ii) a novel context encoder to condition any state-of-the-art time-series generator on arbitrary numbers and combinations of context variables, (iii) a framework for training this context encoder jointly with a time-series generator using an auxiliary context classification loss designed to increase expressivity of context embeddings and improve model performance. We further provide a comprehensive overview of different evaluation metrics for generative time series models. Our results highlight the efficacy of the proposed method in generating realistic household-level electricity consumption data, paving the way for training larger foundation models in the energy domain on synthetic as well as real-world data.

Figures (9)

• Figure 1: CENTS Architecture. Time series context variables are passed to the context encoder, which generates a compressed context embedding. The context encoder is jointly trained with the generator loss and an auxiliary context reconstruction task that reconstructs the household context from the compressed context embedding. Household context is used to perform within group normalization using a parametric normalizer, and the final household context embedding and the normalized time series are passed to the generative model.
• Figure 2: CENTS with Diffusion-TS captures multi-modal distribution of real time series shapes. Shows a comparison between 100 synthetic time series (blue) and the corresponding closest real time series (red) for the same context.
• Figure 3: CENTS with ACGAN is able to produce highly plausible synthetic time series within dataset bounds, but shows lower variance of time series shapes.
• Figure 4: Shows the mean per-timestep kWh difference over the dataset (red), in-context difference (same location and building type; green), and the model's extrapolated shift in consumption (blue) for Diffusion-TS ($\lambda=0.1$) (top) and ACGAN ($\lambda=0.1$) (bottom). The model's extrapolated shift is less smooth because it is the difference between two synthetic time series, whereas the real and context-matched shift are averages computed over the dataset (or a subset comprising many time series).
• Figure 5: TSNE van2008visualizing visualization of synthetic and real time series data for Diffusion-TS ($\lambda=0.1$).