Loading...

Predicting Pore Water Pressure in Masjed Soleyman Dam Using Explainable Artificial Intelligence

Marashi, Amir Arshia | 2025

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 57925 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Jafarzadeh, Fardin
  7. Abstract:
  8. This study aimed to model pore water pressure in embankment dams using advanced machine learning methods, seeking to improve prediction accuracy by integrating diverse temporal and spatial variables. Data from the Masjed Soleyman Dam, covering the years 2003 to 2024, were used. Initially, the data were cleaned and organized, leading to the introduction of six scenarios based on usable data: two scenarios with pressure gauge inputs, two with piezometer inputs, and two with combined pressure gauge–piezometer inputs. Additional scenarios were later introduced to further refine the analysis. The study employed individual models, including Artificial Neural Networks (ANN), Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), Support Vector Machines (SVM), and Echo State Networks (ESN), as well as ensemble models such as ensANN and ensSVM. Additionally, sensitivity analysis of variables and dimensionality reduction techniques, such as Principal Component Analysis (PCA), were utilized to assess input significance and optimize the models. The results demonstrated that ensemble models outperformed individual models, with ensSVM delivering the best results in most scenarios. Furthermore, ensemble models exhibited greater stability than individual models in post-earthquake conditions, effectively mitigating errors caused by sudden fluctuations. Time was identified as the primary factor in predicting the consolidation trend, while data from nearby piezometers and pressure gauges played a crucial role in capturing seasonal and dynamic fluctuations. SHAP analysis further revealed that spatial variables, particularly data from the closest instruments, had the most significant impact on model accuracy. The application of PCA not only reduced model complexity and computational time but also maintained prediction accuracy.This research highlighted that leveraging diverse input variables and advanced machine learning models can significantly enhance the prediction accuracy of pore water pressure in embankment dams. Future studies are recommended to incorporate data preprocessing techniques, more advanced ensemble models, and real-world validation to further improve model reliability and efficiency.
  9. Keywords:
  10. Pore Water Pressure ; Explainable Artificial Intelligence ; Feature Selection ; Masjed Soleyman Embankment Dam ; Long Short Term Memory (LSTM) ; Principal Component Analysis (PCA)

 Digital Object List

 Bookmark

No TOC