DEEP LEARNING-BASED MULTI-OBJECTIVE RESOURCE ALLOCATION FOR INTERDEPENDENT INFRASTRUCTURE NETWORK RESILIENCE. Masters thesis, King Fahd University of Petroleum and Minerals.

PDF (Final Accepted Thesis) Qusai's Final Thesis.pdf - Accepted Version Restricted to Repository staff only until 3 January 2025. Download (8MB)

Arabic Abstract

شبكات البنية التحتية المترابطة هي عمود الحضارات الحديثة، وضرورية لتقديم الخدمات أԽԲساسية. تركز هذه الدراسة على توقع أهمية العقد داخل هذه الشبكات، والتي تعتبر مكونات حاسمة في الحضارة المعاصرة. لتعزيز مرونة وموثوقية البنية التحتية، تستفيد هذه الدراسة من مزيج من نماذج التعلم آԽԲلي ونماذج التعلم العميق وشبكات أԽԲعصاب الرسومية بطريقة جديدة. تستكشف منهجيتنا تطوير وتقييم نماذج تنبؤية، مع التركيز بشكل أساسي على تحسين تخصيص الموارد متعدد أԽԲهداف. على عكس خوارزميات اԽԲنحدار التقليدية، نستفيد من قوة تقنيات التعلم آԽԲلي المتقدمة، بما في ذلك اԽԲنحدار الخطي، اԽԲنحدار الكثير الحدود، مصنف شجرة القرار، المجاور أԽԲقرب ، Kمعزز التدرج، المجاور أԽԲقرب بالنطاق، اԽԲنحدار الجسري البايزي، الغابة العشوائية، وانحدار الشبكة المرنة. ومع ذلك، ما يميز هذا البحث هو تكامل التعلم العميق وهندسة شبكات أԽԲعصاب الرسومية. تخضع هذه النماذج المعقدة لتدريب وتحقق صارم، مظهرة قدراتها اԽԲستثنائية في توقع أهمية العقد مع تقليل تكاليف اԽԲستعادة وزيادة مرونة شبكة البنية التحتية في الوقت نفسه. لتقييم أداء النموذج، نستخدم مجموعة شاملة من المقاييس التقييمية مثل الخطأ المربع المتوسط )،، (MSEمربع- Rالخطأ المربع المتوسط الجذري )، (RMSEوالخطأ المطلق المتوسط ). (MAEمن الجدير بالذكر أن مصنف شجرة القرار، المجاور أԽԲقرب ، Kوالغابة العشوائية يتفوقون بانتظام على نظرائهم، مسجلين أدنى درجات MSEو .RMSE تحمل هذه الدراسة تأثيرات عملية هامة، حيث تكشف عن إمكانيات نماذجنا لتقييم العقد في الوقت الحقيقي، تخصيص الموارد أԽԲمثل، xxتخفيف المخاطر، والتخطيط لՏՄستجابة للطوارئ داخل نسيج معقد لشبكات البنية التحتية المترابطة.

English Abstract

In contrast to traditional regression algorithms, we harness the power of advanced machine learning techniques, including Linear Regression, Polynomial Regression, Decision Tree Regressor, K-Nearest Neighbors Regressor, Gradient Boosting Regressor, Radius Neighbors Regressor, Bayesian Ridge Regressor, Random Forest Regressor, and Elastic Net Regression. However, what sets this research apart is the integration of Deep Learning and Graph Neural Networks architecture. These sophisticated models undergo rigorous training and validation, showcasing their remarkable capabilities in predicting node significance while simultaneously minimizing restoration costs and maximizing infrastructure network resilience. To assess model performance, we employ a comprehensive suite of evaluation metrics such as Mean Squared Error (MSE), R-squared, Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE). Notably, the Decision Tree Regressor, KNN Regressor, and Random Forest Regressor consistently outshine their counterparts, delivering the lowest MSE and RMSE scores. This research holds significant practical implications, as it unveils the potential of our models for real-time node assessment, optimal resource allocation, risk mitigation, and emergency response planning within the complex fabric of Interdependent Infrastructure Networks. Our interdisciplinary approach underscores the importance of collaboration between infrastructure experts and data scientists, emphasizing ethical considerations and model refinement for responsible predictive modeling in critical infrastructure management. This study represents a pioneering effort in advancing node significance prediction within Interdependent Infrastructure Networks. It provides indispensable tools for infrastructure operators and resilience planners, highlighting the pivotal role of data-driven decision support in safeguarding the functionality of Interdependent Infrastructure Networks and ensuring the well-being of modern society. By seamlessly integrating Machine Learning, Deep Learning, and Graph Neural Networks, this research pushes the boundaries of predictive modeling, paving the way for resilient and reliable infrastructure networks in an ever-changing world. Interdependent Infrastructure Networks are the backbone of modern societies, vital for the delivery of essential services. This research centers on predicting the significance of nodes within these networks, which are crucial components of contemporary civilization. To enhance infrastructure resilience and reliability, this study leverages a combination of machine learning models, deep learning models, and graph neural networks in a novel approach. Our methodology explores developing and evaluating predictive models, primarily focusing on enhancing multi-objective resource allocation. In contrast to traditional regression algorithms, we harness the power of advanced machine learning techniques, including Linear Regression, Polynomial Regression, Decision Tree Regressor, K-Nearest Neighbors Regressor, Gradient Boosting Regressor, Radius Neighbors Regressor, Bayesian Ridge Regressor, Random Forest Regressor, and Elastic Net Regression. However, what sets this research apart is the integration of Deep Learning and Graph Neural Networks architecture. These sophisticated models undergo rigorous training and validation, showcasing their remarkable capabilities in predicting node significance while simultaneously minimizing restoration costs and maximizing infrastructure network resilience. To assess model performance, we employ a comprehensive suite of evaluation metrics such as Mean Squared Error (MSE), R-squared, Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE). Notably, the Decision Tree Regressor, KNN Regressor, and Random Forest Regressor consistently outshine their counterparts, delivering the lowest MSE and RMSE scores. This research holds significant practical implications, as it unveils the potential of our models for real-time node assessment, optimal resource allocation, risk mitigation, and emergency response planning within the complex fabric of Interdependent Infrastructure Networks. Our interdisciplinary approach underscores the importance of collaboration between infrastructure experts and data scientists, emphasizing ethical considerations and model refinement for responsible predictive modeling in critical xviiiinfrastructure management. This study represents a pioneering effort in advancing node significance prediction within Interdependent Infrastructure Networks. It provides indispensable tools for infrastructure operators and resilience planners, highlighting the pivotal role of data-driven decision support in safeguarding the functionality of Interdependent Infrastructure Networks and ensuring the well-being of modern society. By seamlessly integrating Machine Learning, Deep Learning, and Graph Neural Networks, this research pushes the boundaries of predictive modeling, paving the way for resilient and reliable infrastructure networks in an ever-changing world.

Item Type:	Thesis (Masters)
Subjects:	Systems Engineering Math
Department:	College of Computing and Mathematics > lndustrial and Systems Engineering
Committee Advisor:	Almoghathawi, Yasser
Committee Members:	Saleh, Haitham and AlGhazi, Anas
Depositing User:	QUSAI KARRAR (g202113550)
Date Deposited:	04 Jan 2024 07:54
Last Modified:	04 Jan 2024 07:54
URI:	http://eprints.kfupm.edu.sa/id/eprint/142726