Hybrid BiLSTM-PSO Approach for Multi-Metering Point Day-Ahead Electrical Load Forecasting

Accurate individual load profile forecasting is essential for utilities to optimize resource allocation. A novel BiLSTM-PSO approach is proposed to address this challenge. By capturing short-term dependencies in load patterns, it significantly improves accuracy for each metering point. Historical data (BATELEC I, 2018–2023) is analyzed using statistical methods like Pearson correlation, cointegration, and Granger causality to identify these dependencies. A Bi-directional Long Short-Term Memory (BiLSTM) network is then optimized by Particle Swarm Optimization (PSO) for more accurate daily load profile forecasting. The BiLSTM-PSO approach achieves significant reductions in error metrics compared to the baseline model: 49.72% (R-squared), 17.95% (MAE), 17.42% (RMSE), and 30.88% (MSE). Outperforming alternatives, the Bi-LSTM approach offers superior forecasting accuracy. The results recommend integrating a BiLSTM-PSO model into information systems for real-time load forecasting, improving procurement planning and grid management efficiency for electric utilities.