| G. Vonitsanos, Ph. Mylonas, A. Kanavos |
| A Comparative Evaluation of Prophet and LSTM Models for Electricity Consumption Forecasting in Smart Meter Time Series |
| 17th International Conference on Information, Intelligence, Systems and Applications (IISA 2026), 6-9 July 2026, Rhodes, Greece |
|
ABSTRACT
|
| Accurate electricity consumption forecasting is essential for efficient energy management and the reliable operation of modern smart grid systems. The widespread deployment of smart meters has enabled the collection of high-resolution consumption data, creating new opportunities for data-driven forecasting approaches that incorporate both temporal and environmental factors. In this study, residential electricity consumption is analyzed using the London Smart Meter dataset, with temperature considered as a key exogenous variable. A comparative evaluation is conducted between an interpretable statistical model (Prophet) and a deep learning approach based on Long Short-Term Memory (LSTM) networks, using real-world time-series data to assess their ability to capture temporal dynamics and generate accurate forecasts. Experimental results show that the LSTM model consistently outperforms Prophet across all evaluation metrics, including MAE, RMSE, MAPE, and R2, demonstrating a superior capability in modeling nonlinear dependencies and short-term variability. In contrast, Prophet effectively captures global trends and seasonal patterns, but exhibits limitations in handling high-frequency fluctuations and abrupt changes in consumption. The findings highlight the importance of incorporating exogenous variables and selecting appropriate modeling approaches based on the characteristics of the data, providing practical insights for forecasting in residential energy systems.
|
| 06 July , 2026 |
| G. Vonitsanos, Ph. Mylonas, A. Kanavos, "A Comparative Evaluation of Prophet and LSTM Models for Electricity Consumption Forecasting in Smart Meter Time Series", 17th International Conference on Information, Intelligence, Systems and Applications (IISA 2026), 6-9 July 2026, Rhodes, Greece |
[
BibTex] [
Print] [
 Back] |
