The rapid growth of Distributed Energy Resources (DERs), including generation, storage, and flexible loads, is transforming electricity distribution networks. While supporting the energy transition, this evolution increases the risk of voltage and thermal limit violations in Low Voltage (LV) and Medium Voltage (MV) systems. Coordinated control of DERs is therefore essential to maintain safe and reliable grid operation. Dynamic Operating Envelopes (DOEs) provide a promising solution by dynamically allocating power capacity to prosumers, enabling secure operation without costly network reinforcements. However, large-scale implementation of DOEs remains challenging. Their deployment relies on reliable and scalable optimization algorithms, as well as accurate forecasts of domestic behavior. Forecasting in distribution systems is particularly difficult due to high levels of uncertainty. My research, funded by the FRS/FNRS project DEFIE in collaboration with the Luxembourg Institute of Science and Technology (LIST), aims to overcome these barriers. I focus on combining Machine Learning and optimization techniques to integrate forecasting directly into the DOE computation pipeline in order to improve the latter.