A Hybrid Data and Knowledge Driven Risk Prediction Method for Distributed Photovoltaic Systems Considering Spatio-Temporal Characteristics of Extreme Rainfalls

This paper proposes a hybrid knowledge-based and data-driven electrical safety risk (ESR) prediction method considering spatio-temporal characteristics of extreme rainfalls to identify distributed photovoltaic systems (DPVSs) with high risks of shutdowns induced by waterlogging. Firstly, a two-dimensional hydrodynamic partial differential model of DPVS waterlogging is formulated to deduce dynamic distributions of inundation depths under temporal-spatial heterogeneity of extreme rainfalls. A fast image segmentation driven risk partitioning algorithm is developed to extract nonuniform spatial distributions and temporal volatility of rainstorms as well as waterlogging for dividing DPVSs into multiple zones with different degrees of ESRs. Then, a knowledge-based analytical approach for leakage currents concerning inundation depths and parasitic capacitance is mathematically presented to reveal the underlying impacts of extreme rainfalls on ESRs of DPVSs. A data-driven spatio-temporal graph convolutional network is implemented to predict inundation depts of DVPSs for improving ESR prediction accuracy with limited extreme rainfall events and observation samples. Furthermore, probability density functions of spatio-temporal ESRs are formed to dynamically quantify ESR degrees triggering shutdowns of DPVSs in different partitioned zones. Finally, simulation results have validated the effectiveness of the proposed method for the spatio-temporal ESR prediction of DPVSs under extreme rainfalls.