Use of Machine Learning Algorithms to Develop Grey-box Model Equivalents of Distribution Networks

As part of the energy transition, the expansion of renewable energies (RES) and the shutdown of conventional, thermal power plants are shifting the focus of generation from the transmission grid to the distribution grids. This changes the power flows between the voltage levels and thus influences the operation and stability of the overlying transmission grid. Due to the high computational cost, distribution networks are often modeled by very simple passive load models as part of transmission system analyses. For distribution grids with a high penetration of RES with a very volatile feed-in behavior, these models need to be questioned. Therefore, within this paper, a grey-box model approach is used in order to develop adequate network equivalents. Two machine learning algorithms, a genetic algorithm, and a shuffled-frog-leaping algorithm are used to estimate the parameters of the grey-box model. The results of the two algorithms are compared based on their solution quality, convergence behavior, and robustness.