Bridging PLEXOS and AI: A Multi-Agent, LLM-Based Framework for Transparent and Compliant Energy Planning

Lodh, Suvajit (2025) Bridging PLEXOS and AI: A Multi-Agent, LLM-Based Framework for Transparent and Compliant Energy Planning. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (4MB) | Preview
Preview	PDF (Configuration Manual) Download (1MB) | Preview

Abstract

PLEXOS is a simulation tool which has been traditionally used in energy infrastructure planning. Although strong, these tools are computationally intensive, opaque, and require expert interpretation, and therefore are limited in their responsiveness to quickly-changing policy and market conditions. As EU AI Act would require high-risk AI systems to be auditable and explainable, new methods that would be simultaneously efficient, explainable, and AI-compliant are of high demand. The idea to combine Large Language Models (LLMs) and energy system simulators to create an AI guidance overlaid over an energy system simulation is developed and explored in the thesis. The framework will automate the scenario, parameter estimation, simulation execution, KPI extraction, and so complete the scenario analysis faster, to increase transparency, and support superior decision-making. The architecture is based on Autogen multi-agent where coding is done specially in simulation management, data analysis, dynamic code execution, and error correction. A governance layer provides a bridge to regulatory compliance by means of audit trails and natural language justifications. Parallel computing is also provided through the use of high-performance facilities to execute a series of analytical workflows. Evaluation indicated that the system was able to perform analytical cycles in a few seconds consistently. Outputs were traceable and transparent and with a rational explanation of the same was improving user trust and readiness to comply. The system is much more efficient, transparent and aligned to regulations when compared to the traditional methods of planning. Keeping within an integrated architecture, it introduces a new standard of explainability and governance meta-mechanisms within AI-driven energy planning. The accuracy of the framework is limited by the accuracy of the underlying simulation models and reasoning of optimal paths that may still yield occasional output errors. Real-time adaption to the market and generalization to a broader domain are open problems of further research work.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Haque, Rejwanul UNSPECIFIED
Subjects:	Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence P Language and Literature > P Philology. Linguistics > Computational linguistics. Natural language processing H Social Sciences > HD Industries. Land use. Labor > Specific Industries > Energy industries
Divisions:	School of Computing > Master of Science in Artificial Intelligence
Depositing User:	Ciara O'Brien
Date Deposited:	04 Jul 2026 13:49
Last Modified:	04 Jul 2026 13:49
URI:	https://norma.ncirl.ie/id/eprint/9472

Actions (login required)

View Item