Improving the Security-Constrained Operative Planning of Flexibilities in Distribution Grids Using Artificial Intelligence and High-Performance Grid Calculation

The penetration from renewable distributed energy resources (DERs) such as wind and solar massively increases the risk of grid congestions on transmission lines/transformers and voltage violations. To guarantee grid security, the operators must perform grid congestion management (GCM), e.g., with active power curtailment from DERs in the operative planning and real-time phases. Compared to the real-time GCM, for grid operative planning, grid congestions must be identified through forecasting-based grid simulations. Distributed flexibility for GCM needs to be planned with, e.g., day-ahead load and DER forecasting. The core challenge of the process is the high computational overhead, which is addressed in this thesis.
This thesis focuses on developing computational tools to improve grid operative planning. Firstly, a high-performance grid simulator with GPU acceleration is developed for the efficient evaluation of many grid statuses. The grid simulator enables the evaluation of the impact of forecasting uncertainties with probabilistic grid simulation. Secondly, a deep reinforcement learning-based Artificial Intelligence (AI) optimization approach for the grid operation is developed. The approach can perform multiple optimizations to reduce the grid operational costs, such as minimization of active power curtailment for GCM. The novelty is that the training of the AI model can be performed self-supervised with the high-performance grid simulator as well as combined with the classical supervised training approach. After training, the method achieves high optimality with significant computational performance improvement over mathematical optimization. Specifically, for the flexibility planning for GCM, the AI optimization approach is extended for active/reactive power (PQ) flexibility area estimation at the transmission system - distribution system interface (extra high voltage - high voltage transformers) with the DER flexibility of the high voltage grid. The optimization considers the robustness against forecasting uncertainties and the realistic N-1 grid security criterion as extended grid-security constraints. High computational efficiency and the simultaneous identification of corresponding DER PQ setpoints in the identified available PQ area are the highlights of the method.
The proposed tools are verified with case studies and find successful applications in multiple projects and research works.