Betreuer/in: Bazan
Überschrift
This master’s thesis investigates whether a GenAI assistant can support energy system simulations by helping users configure parameters and understand results without adding new errors.
Energy simulations are difficult because they require many interdependent inputs, and even small mistakes in battery capacity, PV setup, or tariffs can make a run invalid. Two systems were built and compared using the same technical stack: a multi-agent architecture and a single-prompt baseline.
The multi-agent system uses specialized agents for parameter extraction, field-by-field collection with RAG support, validation through a backend API, and interpretation of simulation outputs. The baseline handles extraction, clarification, and validation through one large prompt with memory, but it only supports configuration tasks. In tests across ten configuration scenarios, the multi-agent system achieved higher completeness, finishing at 98.8% compared with 94.7% for the baseline. It was also faster and more token-efficient, although the baseline required fewer conversation turns overall. The results show that the multi-agent approach is better for complex inputs and non-expert users, while the single-prompt approach can work well when users provide a complete specification at once.
Room 04.137, Martensstr. 3, Erlangen
or
Zoom:
https://fau.zoom-x.de/j/68350702053?pwd=UkF3aXY0QUdjeSsyR0tyRWtLQ0hYUT09
Meeting-ID: 683 5070 2053
Kenncode: 647333