This report compares two control architectures for managing congestion in power grids. The first is a state-of-the-art feedforward controller that relies on a model of the grid and forecasts of possible disturbances to adjust the generation. The other controller is based on Online Feedback Optimization (OFO), and uses measurements to adjust the generation in real-time. This is done by using a projected gradient descent control algorithm with input and output constraints. For the comparison of the two control architectures, a Python library called Pandapower is used, with a 9-bus grid that has 2 PQ generators. The results show that feedforward control is not robust to forecast errors. Meanwhile, feedback control can adjust in real-time and adapt to disturbances. This suggests that feedback control is better suited for power grid congestion management with a large share of renewable generation.
Ahling et al. (Wed,) studied this question.