Coherency Identification and Aggregation in Grid-Forming Droop-Controlled Inverter Networks

There is an increasing need to apply rigorous model-order-reduction techniques in the analysis of large-scale networks of inverter-based distributed generation resources due to the limitations of existing simulation tools. Various coherency-based aggregation techniques have long been used to construct reduced-order dynamic models of large-scale synchronous machine (SM) networks. Such techniques have the advantage of preserving the nonlinear nature of the dynamic model throughout the order-reduction process, enabling the efficient and accurate analysis of large-scale network dynamics during large disturbances such as fault events. This paper proposes the application of a rigorous coherency-based aggregation technique to the analysis of large-scale networks of grid-forming droop-controlled inverters. A rapid and powerful generalized eigenvalue perturbation technique for coherency identification, previously only applied to SM networks, is adapted to grid-forming droop-controlled inverter networks. The resulting reduced-order models are physically insightful and are capable of accurately reproducing the system response in the aftermath of large disturbances. For some networks, a rigorously-derived condition of coherency can be difficult to achieve, given the expected range of L-C-L filter impedances. To remedy this limitation, the potential for high-bandwidth inverter control to enforce the conditions that allow for coherency of droop-controlled inverters has been investigated and confirmed using a controller hardware-in-the loop testbed. Using this approach, the use of simple nonlinear aggregate inverter models to accurately model large sections of the inverter network can be more rigorously justified.