Hendrik Lens

Hassan Abouelgheit, Hendrik Lens

The assessment of reactive power demand plays an instrumental role in power system planning. This paper presents a methodology for calculating reactive power demand based on a two-step approach. Unlike existing methodologies in the literature that focus primarily on optimization of reactive power compensation equipment placement and sizing through single-simulation approaches, this methodology directly calculates the actual reactive power demand through a comprehensive back-to-back simulation framework. While existing methods address either long-term or short-term voltage stability using either steady-state analysis or individual dynamic simulations, the proposed approach integrates both stability assessments sequentially through iterative Quasi-Dynamic Simulation, Q-V analysis and dynamic simulation. Furthermore, this methodology employs comprehensive time-series analysis over a full annual period (8760 hours) with multi-criteria violation assessment (number, severity and duration of voltage violations). In the final section of this paper, a case study was conducted to demonstrate the application of the proposed methodology. Simulations were performed to validate the effectiveness of the methodology, with the results showing that all buses with voltage issues were successfully addressed and finally the total reactive power demand across the network was calculated.

Georgios Mitrentsis, Hendrik Lens

PV power forecasting models are predominantly based on machine learning algorithms which do not provide any insight into or explanation about their predictions (black boxes). Therefore, their direct implementation in environments where transparency is required, and the trust associated with their predictions may be questioned. To this end, we propose a two stage probabilistic forecasting framework able to generate highly accurate, reliable, and sharp forecasts yet offering full transparency on both the point forecasts and the prediction intervals (PIs). In the first stage, we exploit natural gradient boosting (NGBoost) for yielding probabilistic forecasts, while in the second stage, we calculate the Shapley additive explanation (SHAP) values in order to fully comprehend why a prediction was made. To highlight the performance and the applicability of the proposed framework, real data from two PV parks located in Southern Germany are employed. Comparative results with two state-of-the-art algorithms, namely Gaussian process and lower upper bound estimation, manifest a significant increase in the point forecast accuracy and in the overall probabilistic performance. Most importantly, a detailed analysis of the model's complex nonlinear relationships and interaction effects between the various features is presented. This allows interpreting the model, identifying some learned physical properties, explaining individual predictions, reducing the computational requirements for the training without jeopardizing the model accuracy, detecting possible bugs, and gaining trust in the model. Finally, we conclude that the model was able to develop complex nonlinear relationships which follow known physical properties as well as human logic and intuition.