# A World Model Based Reinforcement Learning Architecture for Autonomous Power System Control

Source: KTH DiVA/IEEE metadata and KTH doctoral-thesis summary of included paper; arXiv search returned no arXiv version.
Retrieved: 2026-06-14

## Bibliographic Metadata

- Authors: Magnus Tarle; Mårten Björkman; Mats Larsson; Lars Nordström; Gunnar Ingeström
- Venue/status: 2021 IEEE International Conference on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm), pp. 364-370. DOI 10.1109/SmartGridComm51999.2021.9632332.

## Source Links

- https://doi.org/10.1109/SmartGridComm51999.2021.9632332
- https://kth.diva-portal.org/smash/record.jsf?pid=diva2%3A1637952
- https://dblp.org/rec/conf/smartgridcomm/TarleBLNI21
- https://kth.diva-portal.org/smash/get/diva2%3A1996061/FULLTEXT01.pdf

## Abstract / Retrieval Summary

The paper presents the World Model for Autonomous Power System Control (WMAP), a model-based reinforcement learning architecture that learns an internal world model and includes a safety shield to reduce the risk of poor decisions under high uncertainty. The shield can require human guidance when the agent is unsure or run in decision-support mode. The case study uses IEEE 14-bus power-system control with FACTS setpoints, not Grid2Op topology-control benchmarking.

## Retrieval Note

No arXiv version or open standalone full-text PDF was found. A later KTH doctoral thesis includes a summary and reference to the paper but not the full article body; stored as metadata-only.