Experimental Assessment of LVDC-Grid Stability Optimization using Circuit Simulation and Machine Learning

Low-voltage direct current networks play a central role in the realization of a sustainable, resilient energy supply. The stabilization of the networks during design and operation of the grids is a particular challenge since more and more components must be included in the stability analysis and control. Therefore, it is desirable to increasingly apply automated computations and artificial intelligence techniques for stability optimization. This paper describes the investigations for the experimental validation of a new approach for stability optimization, which can be applied in regular grid operation. A direct current network is mapped into a digital twin for automated computation of stability states. A classification model establishes the relationship between the network input parameters and the stability states, so that improved network parameter settings can be determined with a novel optimization method. A novel impedance measurement based on pseudorandom binary sequences (PRBS) enables precise characterization of grid components and continuous stability monitoring. Following the component characterization, the digital twin model is transferred and validated on a direct current network testbed and at the example of adjusting characteristic droop control curves.