Velvis, Lucas (2025) Hybrid Multi-Agent Reinforcement Learning for Low-Carbon Dispatch in Renewable Energy Valleys. Master's Thesis / Essay, Artificial Intelligence.
|
Text
mAI2025LucasVelvis.pdf Download (2MB) | Preview |
|
![[img]](https://fse.studenttheses.ub.rug.nl/style/images/fileicons/text.png) |
Text
Toestemming.pdf Restricted to Registered users only Download (212kB) |
Abstract
The European Union must cut 55% of 1990-level emissions by 2030. However, intermittent renewables complicate electricity-flow optimization in the recently emerging Renewable Energy Valleys. We present MARLOES, a simulation benchmark for energy hubs, and propose a hybrid multi-agent Dyna-SAC controller which combines model-free robustness with model-based sample efficiency. Over ≈ 35 day simulations, using 20% synthetic rollouts, the algorithm cuts CO₂-equivalent emissions by 32.7% versus a heuristic baseline and 23.4% relative to pure SAC. Under additional sensor and forecast noise the hybrid advantage vanishes, confirming that world-model benefits depend on model reliability. These results show that carefully calibrated model-based augmentation can advance low-carbon dispatch in distributed energy systems aligned with EU decarbonization goals, and that MARLOES provides a flexible platform for future approaches.
| Item Type: | Thesis (Master's Thesis / Essay) |
|---|---|
| Supervisor name: | Grossi, D. |
| Degree programme: | Artificial Intelligence |
| Thesis type: | Master's Thesis / Essay |
| Language: | English |
| Date Deposited: | 25 Jul 2025 06:43 |
| Last Modified: | 25 Jul 2025 06:43 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/36525 |
Actions (login required)
 |
View Item |