An Electric Load Forecasting Enhancement Using LSTM-RNN Integrated with Genetic Algorithms: A Case Study of SEWA in the Sharjah Emirate

Abstract

Introduction: Accurate electric load forecasting is essential for power companies to ensure efficient load scheduling and avoid excessive electricity production. However, developing a reliable forecasting model remains a complex task due to the need for training multiple models, selecting the most effective one, engineering informative features, and determining optimal time lags for time series forecasting.

Objectives: This research aims to address the challenges of long-term electric load forecasting by proposing a hybrid model that combines Long Short-Term Memory (LSTM) neural networks with Genetic Algorithms (GA). The objective is to enhance forecasting accuracy through optimal feature selection, time lag determination, and LSTM architecture configuration.

Methods: The methodology involves building baseline models using both linear and non-linear machine learning algorithms. Feature selection is conducted using wrapper and embedded methods to identify the most relevant predictors. Genetic Algorithms are applied to optimize time lags and the number of LSTM layers. The model is trained and evaluated using electricity consumption data from the Sharjah Emirate.

Results: The optimized LSTM-GA model significantly outperforms traditional machine learning models. It achieved a Mean Absolute Error (MAE) of 10 MW and a Root Mean Square Error (RMSE) of 12 MW, compared to 20 MW MAE and 25 MW RMSE from the Random Forest model. Feature selection reduced MAE by 33.3% and RMSE by 40%, demonstrating the effectiveness of GA-based optimization. Temperature analysis revealed that every 1°C increase beyond 30°C results in a 900 MW rise in electricity load, highlighting the importance of incorporating weather data.

Conclusions: The results confirm that the proposed LSTM-GA framework effectively captures the dynamics of long-term electric load forecasting. The integration of feature engineering, optimization algorithms, and weather data enhances prediction accuracy, making it a promising tool for power system planning and management.