Abstract
In Libya, the general electricity company is tasked with managing peak electricity demand, often resorting to load shedding. This practice, while necessary, results in power outages, particularly impacting areas like the Benghazi Electrical Grid. This study aims to bring predictability to these events by exploring time series forecasting models namely: Autoregressive Integrated Moving Average (ARIMA), Seasonal ARIMA (SARIMA), and Dynamic Regression ARIMA (DRARIMA). The models were trained using data from May 2020 and 2021, and subsequently tested on May 2022. Performance was evaluated using metrics such as mean squared error, mean absolute error, mean absolute percentage error, and mean absolute percentage accuracy. The ARIMA model achieved the highest accuracy at 78.88% mean absolute percentage accuracy with a mean absolute error of 0.9. The SARIMA model, which considers seasonal patterns, achieved an accuracy of 73.86% and mean absolute error of 0.11, but its complexity may lead to overfitting. The DRARIMA, which incorporates exogenous variables, demonstrated an accuracy of 65.36% and mean absolute error of 0.15. Future projections for May 2024 and 2025 using ARIMA models indicate potential improvements in load shedding management and highlight the importance of model selection for accurate forecasting. By improving load forecasting accuracy, this research aims to enhance the effectiveness of load shedding management, thereby reducing power outages and their socio-economic impacts in regions like Benghazi. These findings are particularly valuable for energy planners and managers in similar contexts, providing practical insights and data-driven strategies.

Key words: Autoregressive Integrated Moving Average, Benghazi Electrical Grid, General Electricity Company of Libya, Load Shedding, Time Series Analysis.