Decentralized Data-Driven Tuning of Droop Frequency Controllers applied to a Non-Linear Grid Model

The power grid relies on several controllers to maintain the electricity supply stable and within acceptable quality levels. These control mechanisms are traditionally tuned manually based on models of the current grid configuration. However, manual tuning methods are unable to meet the requirements of modern power grids, where the installation of new generating units and changes in the topology are ubiquitous. In this work, a decentralized automated droop tuning algorithm is analyzed. It utilizes data-based system identification to estimate the remaining grid in the perspective of the local generator/control area, and optimizes the tunable parameters using local and inferred data. The algorithm is deployed and evaluated for a non-linear model of the transmission grid. The proposed method is compared against an all-knowing baseline and a static method, in which the droop values are insensitive to changes in the grid configuration. Three network topologies with increasing number of nodes and connections are used as study scenarios to validate the decentralized algorithm. It is shown that the studied method reduces the variance of the frequency oscillations in the grid by 27% in comparison to the static approach and comes within 3% of the baseline.