[Paper Review] Simulating Tariff Impact in Electrical Energy Consumption Profiles with Conditional Variational Autoencoders

Margaux Brégère, Ricardo J. Bessa|arXiv (Cornell University)|Jun 10, 2020

Smart Grid Energy ManagementEngineering51 references19 citations

TL;DR

This paper proposes a conditional variational autoencoder (CVAE)-based method to simulate household electricity consumption profiles under different tariff schemes, incorporating weather and calendar variables. Unlike semi-parametric models, the CVAE captures non-linear rebound and side effects—where consumption shifts before or after tariff changes—enabling realistic ex-ante testing of demand response policies across consumer clusters with varying price responsiveness.

ABSTRACT

The implementation of efficient demand response (DR) programs for household electricity consumption would benefit from data-driven methods capable of simulating the impact of different tariffs schemes. This paper proposes a novel method based on conditional variational autoencoders (CVAE) to generate, from an electricity tariff profile combined with exogenous weather and calendar variables, daily consumption profiles of consumers segmented in different clusters. First, a large set of consumers is gathered into clusters according to their consumption behavior and price-responsiveness. The clustering method is based on a causality model that measures the effect of a specific tariff on the consumption level. Then, daily electrical energy consumption profiles are generated for each cluster with CVAE. This non-parametric approach is compared to a semi-parametric data generator based on generalized additive models and that uses prior knowledge of energy consumption. Experiments in a publicly available data set show that, the proposed method presents comparable performance to the semi-parametric one when it comes to generating the average value of the original data. The main contribution from this new method is the capacity to reproduce rebound and side effects in the generated consumption profiles. Indeed, the application of a special electricity tariff over a time window may also affect consumption outside this time window. Another contribution is that the clustering approach segments consumers according to their daily consumption profile and elasticity to tariff changes. These two results combined are very relevant for an ex-ante testing of future DR policies by system operators, retailers and energy regulators.

Motivation & Objective

• To develop a data-driven, non-parametric method for simulating household electricity consumption profiles under various tariff schemes.
• To enable ex-ante evaluation of demand response (DR) policies by generating realistic consumption data conditioned on tariffs, weather, and calendar variables.
• To cluster consumers based on their tariff responsiveness using a causality model, allowing personalized DR signal design.
• To overcome limitations of semi-parametric models that fail to capture temporal spillover effects like rebound and side effects in consumption behavior.
• To provide a reproducible, open-source simulation framework for energy regulators, retailers, and system operators to test future DR programs.

Proposed method

• Consumer households are clustered into k groups (C₁,…,Cₖ) based on their consumption patterns and responsiveness to tariff changes using a causality model that quantifies the causal effect of tariffs on consumption.
• A conditional variational autoencoder (CVAE) is trained to generate daily half-hourly consumption profiles Y, conditioned on input variables x (tariff p, weather, calendar type w), with latent space Z of dimension d.
• The CVAE decoder generates consumption profiles by mapping the latent variable Z and conditional inputs (p, w, κ) to a distribution over Y, modeling complex, non-linear dependencies.
• The model is trained on a public dataset from UK Power Networks, using a variational inference objective to maximize the lower bound of the log-likelihood of observed consumption.
• The CVAE-based generator is benchmarked against a semi-parametric model based on generalized additive models (GAMs), which uses prior knowledge of energy consumption behavior.
• The method supports conditional generation of consumption profiles for unseen tariff profiles, though generalization is limited when inputs fall outside the training distribution.

Experimental results

Research questions

• RQ1 Can a CVAE-based generator accurately simulate household electricity consumption profiles under different tariff schemes, including those not explicitly observed during training?
• RQ2 How well does the CVAE model capture rebound and side effects—where consumption shifts before or after a tariff change—compared to semi-parametric models?
• RQ3 To what extent does the clustering of consumers based on tariff responsiveness improve the realism and diversity of generated consumption profiles?
• RQ4 Does the non-parametric CVAE approach outperform or match the performance of semi-parametric models in reproducing the mean consumption values of the original data?
• RQ5 How does the model’s performance vary across different consumer clusters with distinct price elasticity and behavioral patterns?

Key findings

• The CVAE-based generator achieves performance comparable to the semi-parametric benchmark in reproducing the mean consumption values of the original data.
• The CVAE model successfully captures rebound effects—such as increased afternoon consumption when a High tariff is applied in the evening—and side effects, where consumption changes persist before and after the tariff window.
• Unlike the semi-parametric model, which only affects consumption during the exact half-hours when tariffs change, the CVAE model detects and reproduces spillover effects across time, reflecting real consumer behavior.
• The clustering approach effectively segments consumers based on their tariff responsiveness, with distinct rebound and side effect magnitudes observed across clusters.
• The CVAE generator produces non-intuitive profiles when given rare or unseen tariff inputs (e.g., a full day of High tariff), indicating limited generalization capacity outside the training distribution.
• The semi-parametric model generates unreasonably low consumption during full-day High tariffs, suggesting it fails to model the physical and behavioral limits of energy use.