The purpose of this study is to address the critical challenge of short-term electricity demand forecasting in Morocco, where climate variability increasingly affects energy consumption. This study proposes a hybrid forecasting framework that explicitly incorporates average temperature as a climatic driver to improve forecast accuracy and inform energy planning and policy.
This study develops a hybrid model combining a Seasonal Autoregressive Integrated Moving Average with Exogenous Variables and a Multilayer Perceptron (MLP) neural network. This architecture captures both linear seasonal trends and nonlinear residual patterns in electricity demand. The model is trained on monthly data from January 2008 to March 2020, with average temperature included as a key exogenous input.
The hybrid model consistently outperforms standard benchmarks, delivering higher accuracy over a 29-month forecast horizon. It successfully predicts demand surges during simulated heatwaves, demonstrating its capacity to capture climate-sensitive consumption behavior. These results have practical applications for infrastructure planning, grid reliability and seasonal maintenance.
The model’s accuracy is constrained by data limitations and the exclusion of other relevant drivers such as policy shifts and behavioral responses. Future work could extend the framework by incorporating additional socioeconomic and regulatory variables.
This paper offers a novel application of hybrid time series and machine learning methods to the Moroccan energy context. By integrating temperature as a climatic input, the model enhances predictive performance under extreme weather conditions. This contributes to a more climate-resilient forecasting strategy and provides a replicable framework for other regions facing similar challenges.
