Post-fault voltage recovery and voltage instability assessment of DC microgrid with Deep Transfer-learning Convolution Neural Network

Abstract

Extreme events lead to undesirable scenarios in a Distributed Energy Resources (DER) integrated DC microgrid. On higher penetration of renewable energy resources like Solar Photovoltaic Systems (PV's), and wind and battery energy storage systems, the instability in the microgrid tends to increase due to the presence of converter dynamics during faults and sudden changes of loads. This voltage instability becomes inherent in a highly renewable energy source penetrated DC microgrid system since the system operations rely entirely on the converters. Since the DC microgrid is mainly connected to the distribution system, it is also important to maintain the nominal voltage at $\pm 10\%$ according EN 50155 standard. This paper proposes a deep learning-based post-fault voltage recovery and voltage instability assessment to reconnect the DC microgrid. In this research, a Deep transfer learning-based Convolution Neural Network (DTCNN) is adapted for the first time, which uses the features extracted from raw time-series data converted into spectrums with small samples, then pre-trained for online assessment for voltage instability in a DC microgrid. The proposed mechanism performs a visualization of a high dimensional data classification using T-distributed Stochastic Neighbourhood Embedding (t-SNE) to correlate between high dimensional data features and improve the generalized performance of the DTCNN. If the voltage stability indicator is a non-zero value, then the DC microgrid will likely perform event-triggered preventive control or remedial actions by disconnecting DERs/VSCs or balancing power through a readily dispatchable Energy Storage System. Extensive Time Domain Simulation (TDS) for post-fault transient response is performed in the MATLAB/Simulink platform®. The results provide insight into the post-fault recovery and voltage instability assessment for the DC microgrid subject to fault and islanding events.