Hybrid Artificial Bee Colony Long Short-Term Memory Networks for Enhanced Short-Term Load Forecasting

Abstract

Short-term load forecasting is critical for optimising energy distribution in urban networks, yet challenges arise from non-linear and non-stationary demand patterns. This study addresses the problem of inaccurate load predictions in the Abuja Electricity Distribution Company (AEDC) network, which can lead to inefficient resource allocation and grid instability. The aim was to develop a hybrid Artificial Bee Colony (ABC)-Long Short-Term Memory (LSTM) model to enhance forecasting accuracy across three distribution areas: Lugbe, Gwagwalada, and Abaji. The methodology integrated ABC for hyperparameter optimisation with LSTM for temporal dependency modelling, using hourly load data from 2020–2024. Results demonstrated a consistent 94.0% peak load accuracy across all areas, outperforming traditional LSTM (90.2%) and ensemble methods (88.7–91.3%). Hour accuracy varied from 17.99% to 41.48%, indicating temporal prediction challenges. The model’s robust generalisation and stable forecasts, confirm its utility for operational planning. This hybrid approach advances accurate and adaptable load forecasting, offering a scalable solution for intelligent energy management.