Enabling Decision Insight by Applying Monte Carlo Simulations and Eigenvalue Spectral Analysis to the Ship-Centric Markov Decision Process Framework.

Ship design;decision making;Markov decision process;eigenvalue analysis;Naval Architecture and Marine Engineering;Engineering;Naval Architecture and Marine Engineering

Kana, Austin A.Alford, Laura Kay ;

University of Michigan

关键词: Ship design;    decision making;    Markov decision process;    eigenvalue analysis;    Naval Architecture and Marine Engineering;    Engineering;    Naval Architecture and Marine Engineering;   

Others  :  https://deepblue.lib.umich.edu/bitstream/handle/2027.42/120673/akana_1.pdf?sequence=1&isAllowed=y

瑞士|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

One of the major problems facing ship design today is that engineers often focus most of their efforts on the What of the design as opposed to understanding the Why. The What is defined as the solution itself, while the Why is defined as the decisions that drive how the set of solutions change through time. Decision making through time, especially in the face of uncertainty, has consistently been difficult for engineers. This is due to both uncertainty and the interconnected nature of complex decision making problems. There are no standard definitions or metrics that quantify the impact of engineering design decisions. This dissertation aims to address that need. This research extends the ship-centric Markov decision process (SC-MDP) framework which involves applying Markov decision processes to ship design and decision making. The SC-MDP framework is useful for analyzing decision making in the maritime domain due to its inherent temporal structure and ability to handle uncertainty. However, the framework is limited in its ability to clearly show how uncertainty affects decisions, and its inability to quantify the changes and long term implications of decisions.Two methods unique to this research are developed and explored. First, applying Monte Carlo simulations to the SC-MDP framework is proposed to give insight into the impacts of uncertainty on the decisions and set of results. Second, a method to perform eigenvalue spectral analysis within the framework was developed to understand the behavior of the decisions themselves. Three metrics are developed in regards to eigenvalue analysis. To quantify changes in decisions, the damping ratio is proposed, defined as the ratio of the largest eigenvalue to the magnitude of the second largest. To understand the long term implications of a set of decisions the principal eigenvector is presented. For eliciting relationships and inter-dependencies of decisions, analyzing repeated dominant eigenvalues and the set of principal eigenvectors are used. Three maritime case studies are presented that demonstrate the utility of these methods and metrics involving designing for evolving Emission Control Area regulations, ship egress analysis and general arrangements design, and lifecycle planning for ballast water treatment regulations.