Robust Optimization Approach for Systems under Uncertainty via Hybrid Metamodeling

This dissertation explores the application of advanced machine learning techniques, specifically deep learning (DL) and deep reinforcement learning (DRL), to solve complex optimization problems in various domains. The research begins with robust simulation optimization in supply chain management under uncertainty using neural network metamodeling, establishing a foundation for understanding complex system behaviors. Building on this, the study addresses maintenance policy optimization in lean production systems through DRL, demonstrating the adaptability and efficiency of these techniques in dynamic environments. Finally, the dissertation applies DRL to the AC Optimal Power Flow (AC-OPF) problem, developing a modified warm-start approach that integrates Q-learning and a novel loss function based on the probability distribution of solution input decision variables. While achieving improvements in solution quality and cost-effectiveness, the method presents mixed results regarding computational speed. This work highlights the potential of DL and DRL in optimizing operations across diverse fields while also identifying avenues for future research to enhance computational efficiency.