A scalable and resilient protection framework for hybrid microgrids using zero injection cluster and graph learning

Hybrid microgrids in spite of offering a promising solution to meet rising energy demands, have not received wider acceptance by power utilities because of the complexity of their protection schemes. Real-world microgrids are highly susceptible to disruptions, during extreme weather conditions resulting in frequent line outages and sensor failures. Further complications arise from the variations in operational dynamics caused by weather dependent intermittent behavior of solar and wind distributed energy resources (DERs). Failing to address these issues, hinders accurate fault detection/classification under extreme weather conditions, thereby impacting the microgrid resilience. In this regard, a protection framework using zero injection cluster (ZIC) and graph learning with resilience against contingency scenarios and weather intermittency is proposed for the hybrid microgrid. The present work incorporates the effect of ZIC to formulate the critical sensor identification problem with the aim of minimizing sensor installation costs while enhancing measurement redundancy. The same imparts scalability to the protection scheme with regard to the architecture and size of the microgrid. To accommodate intermittency and potential correlations between solar and wind DERs, a joint probabilistic approach, encompassing the uncertainty present in both sources is considered. This work employs a spatiotemporal graph convolutional network classifier to detect and classify faults by integrating the network topology information into the protection framework. Validation of the proposed scheme for varying fault and operating scenarios reveals its ability to attain high degree of accuracy in fault detection and classification with increased resilience and immunity. • Scalable protection framework for hybrid microgrids using graph learning approach. • Critical sensor identification considering the effect of zero injection cluster. • Modeling of solar and wind intermittency using joint probabilistic approach. • Resilience against N-1 contingency scenarios caused by extreme weather events. • Spatiotemporal graph convolutional network classifier modules for protection tasks.