Thèses de doctorat

Permanent URI for this collection

https://depo.univ-annaba.dz//handle/123456789/113

Browse

Browsing Thèses de doctorat by Author "BOGHANDJIOUA, Samira"

Now showing 1 - 1 of 1
  • No Thumbnail Available
    Item
    Etude d'impact du changement climatique sur la sensibilité hydraulique des systèmes de drainage urbain
    (Université Badji Mokhtar Annaba, 2025) BOGHANDJIOUA, Samira
    Floods are a major threat to urban areas in Algeria, as rapid urbanization and climate change are factors that increase the frequency and intensity of rainfall, making drainage systems unable to absorb runoff, posing significant challenges for stormwater management. This thesis studies the impact of these factors on the drainage system of the urban area called Bir Farina in Azzaba city, Skikda province in eastern Algeria, focusing on the limitations of the traditional MIKE+ hydraulic model for the accurate prediction of stormwater network overflows. This thesis describes an innovative approach called SWN-ML (Storm Water Network - Machine Learning), which combines the hydraulic simulations of the MIKE+ model with advanced machine learning algorithms, including ensemble learning techniques, such as Gradient Boosting and Random Forests. A comprehensive database encompassing the geographical, climatic and geometric characteristics of the stormwater network in the study region was developed. The MIKE+ model was calibrated based on the unique measurement of water levels at the stormwater network outlet during the rainfall event of February 4, 2019, and the SWN-ML approach was used to predict the average overflow rates for different rainfall durations and return periods. The results indicate the performance of ensemble learning models (ELM) compared to classical machine learning models in predicting overflow rates. A strong correlation was observed between the predictions of the ensemble methods and the MIKE+ simulations, confirming their effectiveness in capturing the dynamics of the stormwater network. In addition, a feature importance analysis based on Mean Square Error (MSE) identified the main variables influencing overflow rates, providing valuable insights to improve stormwater management strategies. This research highlights the potential of integrating physically based models with machine learning techniques to improve the prediction and management of stormwater network overflows in urban areas. The SWN-ML approach provides a robust and reliable framework for flood riskassessment, enabling the development of effective early warning systems and flood mitigation strategies, crucial for building resilient urban environments.