Improving short‑term electricity load forecasts in Peninsular Malaysia using a hybrid neural network-genetic algorithm approach

The safe and cost-effective operation of contemporary power systems depends on accurate short-term load forecasting, especially in areas where the demand for electricity is rising and renewable energy is becoming more integrated. This study develops and evaluates a hybrid Neural Network–Genetic Algorithm (NN–GA) model for short-term electricity load forecasting in Peninsular Malaysia using historical load data from the national Grid System Operator (GSO) covering June 2023 to October 2024. Correlation analysis between the target load and seven daily lag variables identified the strongest dependency at a one-week lag (Dd–7, r = 0.84), which was therefore adopted as the primary predictor within a three-week lag structure. To capture the weekly load cycle, three representative days were modeled: Monday (start-up phase), Friday (stable mid-week phase), and Saturday (ramp-down phase). The dataset was split chronologically into training (60%), validation (20%), and testing (20%) sets, which corresponded to June-August, September, and October of 2024, respectively. Key neural network hyperparameters, such as the number of hidden neurons, learning rate, and momentum coefficient, were optimized using a Genetic Algorithm (GA). Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE), and Root Mean Square Error (RMSE) were used to assess the model's performance. Results on the October 2024 test set show that the hybrid NN–GA model consistently outperformed or matched standalone NN models, achieving MAPE values of 6.93% (Monday), 2.17% (Friday), and 2.01% (Saturday), with coefficient of determination (R²) values up to 0.93. These findings demonstrate that the NN–GA approach provides a robust data-driven tool for short-term load forecasting, supporting operational planning, peak demand management, and renewable energy integration in Peninsular Malaysia.