Deep Learning-Based Electrical Load Forecasting for Optimized Energy Demand Prediction in the Luzon Grid, Philippines

Short-Term load forecasting (STLF) is critical for ensuring the stability and efficiency of power systems, yet there has been limited research on load forecasting in the Philippines, especially concerning the Luzon grid. Traditional forecasting models have struggled to address the complexities of short-term electricity demand prediction. This study developed a deep learning-based STLF model utilizing Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), and Multilayer Perceptron (MLP) techniques. These models were trained on historical data from 2010 to 2019 and tested on data from 2021 to 2023 to evaluate their robustness and real-world applicability. The study found that temperature, relative humidity, rainfall, weekend or weekday, type of day, month and windspeed significantly influenced load demand, with temperature being the strongest predictor. Among the deep learning models, MLP outperformed both CNN and LSTM, achieving the lowest mean absolute percentage error (MAPE), demonstrating the best performance for forecasting. The MLP model exhibited consistent accuracy across all testing periods with % MAPE of 0.898%, 1.040%, and 1.386% for the years 2021, 2022, and 2023, respectively making it the most reliable model for short-term load forecasting for the Luzon Grid.