Hülsmann, Jonas (2024)
Aspects of Explanations for Optimization-Based Energy System Models.
Technische Universität Darmstadt
doi: 10.26083/tuprints-00027806
Dissertation, Erstveröffentlichung, Verlagsversion

Kurzbeschreibung (Abstract)

The transition towards a renewable energy system presents numerous challenges, demanding extensive decision-making processes. Optimization-based energy system models (ESMs) are valuable tools to facilitate these decisions. However, these models often prove too intricate and expansive for decision-makers, such as CEOs, politicians, or citizens, who typically lack expertise in ESMs. Consequently, there is a pressing need for explanations to bridge the gap between the complexity of ESM results and decision-makers comprehension, allowing them to discern potential disparities between their objectives and the model’s assumptions. This thesis focuses on various explanations for optimization-based ESMs. Initially, we explore explanations from psychological and philosophical perspectives to identify the fundamental concepts necessary for creating comprehensive explanations and elucidate the key factors essential for these explanations to be deemed adequate. Armed with these foundational concepts, we assess the current state-of-the-art in explaining ESMs, identifying existing shortcomings in explanation methodologies. We draw parallels between the challenges faced by the ESM domain and the field of machine learning, particularly in the domain of explainable artificial intelligence (XAI), which is dedicated to developing methods to enhance the explainability of machine learning models. This thesis delves into three distinct approaches from the abovementioned domains to address these challenges and enhance explanations for optimization-based ESMs. Firstly, we adopt an approach from XAI to ESMs to overcome two prevalent shortcomings in ESM explanations: elucidating the impact of high-dimensional input data and tailoring explanation complexities to different target audiences. Given the key role of causality in crafting explanations, we explore the utilization of causal graphical models for explaining ESMs. Finally, we design an interactive approach to facilitate hands-on learning of ESMs, focusing on the example of energy transition, and evaluate the efficacy of this approach within the context of a graduate-level university course.

Frage zum Eintrag

Redaktionelle Details anzeigen