Increasing penetration of Distributed Energy Resources (DERs) in Low-Voltage (LV) grids necessitates novel, reliable control strategies combining grid monitoring and automated decision-making. However, this chained control task poses a significant challenge to Distribution System Operators (DSOs) due to the lack of accurate electrical grid models and automation infrastructure at the LV level. This thesis presents a model-free, data-driven control algorithm for voltage violation mitigation in LV grids, without requiring prior knowledge of physical models or expensive Real-Time (RT) measurements. The algorithm relies on RT measurements and historical Smart-Meter (SM) data to operate the grid within its limits. The method involves a model-free, data-driven State Estimation (SE) and an online-trained Reinforcement Learning (RL) agent. Based on estimated voltages, the RL agent decides on appropriate power factors and curtailment signals for dispatchable DERs to mitigate voltage violations. Simulation experiments show that the controller reduces voltage violations by 98% compared to two baseline algorithms and can handle uncertainties arising from SE deviations. The performance and stability of the controller is improved by extending the RL agent’s observation space with multiple time steps using onedimensional (1D) Convolutional Neural Networks (CNNs). Additionally, this work shows the effectiveness of the controller in low SM penetration scenarios and highlights the importance of SM locations. Overall, this thesis demonstrates that controlling LV grids reliably and cost-effectively with model-free, data-driven methods is feasible.     «

Increasing penetration of Distributed Energy Resources (DERs) in Low-Voltage (LV) grids necessitates novel, reliable control strategies combining grid monitoring and automated decision-making. However, this chained control task poses a significant challenge to Distribution System Operators (DSOs) due to the lack of accurate electrical grid models and automation infrastructure at the LV level. This thesis presents a model-free, data-driven control algorithm for voltage violation mitigation in LV...     »