Analyzing and predicting residential electricity consumption using smart meter data: A copula-based approach

Abstract

Accurate demand prediction is essential for smart grid applications, and its precision can be significantly improved by accounting for individual consumption patterns in smart meter data. As nations and corporations increasingly strive for environmental sustainability, integrating clustering methodologies with forecasting models enables the identification of consumption trends and enhances predictive accuracy. Unlike existing prediction methods focusing on point estimates, we propose a novel clustering-based D-Vine Copula Quantile Regression (DVQR) framework for smart meter demand forecasting, which can capture the distribution of consumption behaviors about external factors such as weather conditions and time of day. The K-means are used to group the residential energy data into different groups. By integrating segmentation techniques with predictive models, DVQR leverages clustering to uncover complex and latent patterns in the data. Furthermore, DVQR extends beyond traditional forecasting by using quantile regression to capture variability, heteroscedasticity, and dependencies in consumption patterns, providing more comprehensive insights into the drivers of electricity demand. Our proposed approach is validated on the Melbourne household's dataset and compared with six models to demonstrate its superior performance. The results show that DVQR offers more accurate and flexible quantile predictions, especially when capturing consumption variability under different conditions.