Thesis

Explainable AI in satellite scheduling : integrating large language models, knowledge graphs, and computational argumentation for onboard and ground segment systems

Downloadable Content

Download PDF
Creator
Rights statement
Awarding institution
  • University of Strathclyde
Date of award
  • 2026
Thesis identifier
  • T17634
Person Identifier (Local)
  • 202050889
Qualification Level
Qualification Name
Department, School or Faculty
Abstract
  • Scheduling satellite operations is crucial for meeting mission requirements and objectives, but as the demand for larger constellations or more advanced onboard or groundbased capabilities grows, traditional management practices are becoming inadequate. To meet emerging needs, there is increasing motivation to automate systems and processes to leverage the latest Artificial Intelligence (AI) technology; however, the nature of space operations necessitates absolute trust in calculations and decisions. The hidden reasoning of AI systems makes it difficult to validate such solutions as satellite scheduling, reducing the opportunity for highly automated and capable systems. A new field of AI study, eXplainable Artificial Intelligence (XAI), aims at solving this problem by providing automated systems with the ability to communicate explanations to users so insight into the decision-making process is possible. Through integrating with systems and models, XAI techniques assist with validating or correcting system behaviour, which can substantiate trust and grant assurances that training and model configuration are functioning as designed. Various approaches for XAI can generate textual and visual-based explanations, utilising argumentation and Knowledge Graph (KG) concepts to quantify explanations and ensure good interpretability for users. This research introduces several novel approaches to meeting XAI requirements, establishing model-agnostic benchmarks in explainability for scheduling systems through the integration of Large Language Model (LLM)s, KGs, and Argumentation Framework (AF)s. An example satellite schedule was derived using a Constraint Programming (CP) Solver to facilitate experimentation of targeting and replanning capabilities, automated tabular data extraction from a database, and query generation and answering to produce explanations for scheduling decisions with all data made available for reproduction. The results from experiments were evaluated for performance in response accuracy, comprehension, and language quality to demonstrate the current capabilities for generating explanations without dedicated model pre-training or extensive optimisation techniques. This thesis concludes with summarising the findings and proposing future research opportunities to leverage the benchmark approaches established here.
Advisor / supervisor
  • Riccardi, Annalisa
Resource Type
DOI

Relations

Items

Thumbnail Title Date Uploaded Visibility Actions
file details PDF of Thesis T17634 2026-03-18 Public Download