Unveiling the Potential of Transformer-Based Models for Efficient Time-Series Energy Forecasting

Abstract

Accurately forecasting energy consumption is critical in optimizing energy management, reducing costs, and enhancing grid stability.This study uses smart meter data to evaluate the performance of four transformer-based models-Vanilla Transformer, Autoformer, Informer, and SpaceTimeFormer-for energy consumption forecasting.The models are evaluated against statistical benchmarks, with results indicating that Autoformer is the most efficient transformer, achieving the best balance between accuracy and computational complexity, with a Mean Absolute Error (MAE) of 0.540, a Root Mean Square Error (RMSE) of 0.764, a Mean Absolute Percentage Error (MAPE) of 0.091, and an R of 0.979.The study focuses on transformer models, establishing their utility for time-series forecasting and identifying Autoformer as the most suitable for this dataset.These findings highlight the transformative potential of advanced architectures for handling complex temporal data and provide a benchmark for future research in energy consumption forecasting.