Analysis and Improvement of LVDC-Grid Stability using Circuit Simulation and Machine Learning - A Case Study

Due to the increasing complexity of Low-Voltage DC microgrid networks, the optimization of stability has become increasingly important in the last decade. Although various techniques for stability assessment and improvement exist, there are few approaches to combine them with automated computational simulation models and machine learning methods to investigate a variety of network parameterizations. The data derived from these simulation experiments enables the establishment of machine learning models on which a systematic optimization of network stability can subsequently be based. Another important prerequisite for stability optimization is the possibility to measure grid stability during operation in comparison with the derived model. In this paper, a novel approach towards the calculation and measurement of grid stability is presented. The grid stability is calculated based on circuit simulations and small-signal analysis applying the minor loop gain criterion. A surrogate model based on machine learning by random forests is developed, which enables rapid prediction of grid stability and analysis of input parameter influence. The capability of measuring impedance by a pseudorandom binary sequence measurement system as an equivalent to small signal analysis enables the transfer of the concept to real-word applications.