Reliability Evaluation of Park-Level Electricity-Hydrogen Systems Using Explainable Graph Neural Network

Integrating hydrogen into electricity systems has been regarded as a promising way to promote sustainable developments. However, traditional methods of reliability evaluation are time-consuming and existing machine-learning-based approaches are lack of transparency. Therefore, in this study, an explainable graph neural network (GNN) is proposed to achieve fast and explainable reliability evaluation of park-level electricity-hydrogen system (PEHS). Specifically, graph convolutional layers are adopted to capture the spreading influence of components based on the connection structure of PEHS. The training and testing data are generated using the proposed Monte-Carlo-simulation-based method. A feature selection algorithm is proposed to provide local explanations which show how important the components are for the evaluation results. Simulation studies are conducted on a revised IEEE 33-bus system. The results verify the effectiveness of the explainable GNN. In addition, the local explanations can show which components are important for the evaluation results.