A heuristic solution approach for dynamic mission abort problem based on deep reinforcement learning

Date

2024-12-5

Author

Yeşiltepe, Duygu

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

53
views

45
downloads

Cite This

This thesis examines the Mission Abort Problem using Deep Reinforcement Learning (DRL) within an Actor-Critic framework, marking a significant departure from traditional stochastic methods for solving high-dimensional Markov Decision Processes (MDPs). The study introduces innovative techniques, such as an instructive pattern strategy to enhance training efficiency, action masking to enforce repair constraints, and the incorporation of the Gamma Process proposed by De Jonge to model system degradation and discretize states. A comparative analysis with classical dynamic programming underscores the capability of DRL to tackle complex decision-making challenges in mission-critical environments.

Subject Keywords

Mission abort problem, Deep reinforcement learning (DRL), Actor-critic network, High-dimensional Markov decision processes (MDPs), Action masking, Degradation modeling, Gamma process, Discretization, Repair, Dynamic optimization

URI

https://hdl.handle.net/11511/113035

Collections

Graduate School of Natural and Applied Sciences, Thesis