Prédiction de la consommation d’énergie électrique dans les bâtiments intelligents

Abstract

Anticipating future electricity consumption is no longer a mere strategic advantage, but a pressing necessity — especially in Algeria, where more than 98% of electricity production relies on fossil fuels, mainly natural gas. In a context where electricity cannot be stored, extreme demand peaks, often caused by climatic events, regularly exceed production capacity. This puts the national grid, managed by Sonelgaz, under severe stress, resulting in frequent outages and compromising overall system stability. This study presents a short-term electricity load forecasting model specifically adapted to Algerian buildings. The proposed approach is based on time series modeling using artificial intelligence, combining Convolutional Neural Networks (CNN) and Long Short-Term Memory networks (LSTM) to capture both local patterns and temporal dependencies in the data. The results obtained are promising : the hybrid CNN-LSTM model achieved a coefficient of determination (R2) of 0.95, a Mean Absolute Error (MAE) of 232, and a Root Mean Squared Error (RMSE) of 330. These metrics confirm the model’s ability to accurately replicate actual consumption and support its potential for operational energy management applications