This thesis analyzes the implementation of a DT model for ANM in power grids, focusing on active network management with intermittent renewable energy sources. Considering the increasing implementation of renewable sources and DES systems, efficient and robust algorithms are important for optimal grid management. We start by evaluating two state-of-the-art RL models, PPO and SAC, as a baseline for expected performance. The SAC model showed superior performance and was used as the teacher model for dataset generation. This dataset was used in the training of the DT model. Our methodology additionally includes PSO to fine-tune the output of our DT. This approach utilizes PSO's ability to understand non-differentiable problem spaces, complementing the DT's predictive accuracy. We evaluate the performance of the DT and the enhanced DT + PSO model in various grid scenarios, especially focusing on the model's behavior during the critical transitional periods between different stages of power demand and generation. Key findings indicate that while the DT aligns closely with the baseline models in terms of performance, it has problems adapting to the sudden shifts between stages A and B in the grid simulation. However, the integration of PSO fine-tuning shows a slight performance improvement, demonstrating the potential of this hybrid approach in improving decision-making in grid management. This thesis contributes to the field by not only introducing a new approach for ANM but also highlighting the importance of model adaptability in various grid scenarios. It lays the groundwork for future research in implementing advanced machine learning techniques for efficient power grid management, especially now as the world transitions towards more sustainable energy systems.