Enhancing long short-term memory for electric load forecasting with multi-batch Bayesian optimization

Bin Chai

doi:10.1016/j.epsr.2025.112436

Enhancing long short-term memory for electric load forecasting with multi-batch Bayesian optimization

2025-10-28

0

PRE

AI

DOI：10.1016/j.epsr.2025.112436

Original

Original request for help

Save

Abstract

En 中文

• An enhanced Long Short-Term Memory model optimized via multi-batch Bayesian optimization method is proposed to improve short-term electricity load forecasting accuracy under complex feature fusion scenarios. • Multivariate time-series inputs are formed by combining historical load and meteorological data to capture real operating conditions. • The model architecture is improved to better capture long-term dependencies, while a hybrid training scheme combining learning rate decay and early stopping, together with optimized hyperparameters, boosts stability, efficiency, and robustness. • Real-world data from different seasons validated the model’s performance.

Keywords:

Short-term load forecasting

Long short-term memory network

Bayesian optimization

Meteorological feature modelling

Enhancing long short-term memory for electric load forecasting with multi-batch Bayesian optimization

Abstract

Journal

Organization