Optimized management of an active distribution network using AMAS combined with the RL bandit method

Modern electrical power systems are evolving with the introduction of distributed energy resources and electric vehicles, promising sustainability. However, the uncontrolled integration of these technologies into legacy power grids can lead to real-time imbalances and peak load issues. Traditional grid reinforcement has drawbacks, including cost and deployment time concerns. Flexible solutions, enabled by grid digitization, offer an alternative by dynamically controlling grid elements. Yet, optimizing these solutions for diverse market actors is complex, and centralized approaches may struggle to manage large-scale smart grids in real-time. This thesis addresses these challenges by developing a decentralized system using adaptive multiagent systems for real-time control of flexible entities in distribution grids. Simulation experiments validate its effectiveness in overcoming centralization issues. Furthermore, integrating combinatorial multi-armed bandit learning enhances performance in stochastic environments. This research offers a promising approach to optimizing large-scale smart grids as they adapt to evolving energy landscapes.