Risk Assessment for Power Grid Infrastructure Projects based on Graph Attention Networks

Current risk assessment methods of power grid infrastructure projects are limited to the post-incident stage, while less research has been conducted on the pre-construction risk assessment for power grid infrastructure projects. To address this limitation, this paper proposed a pre-construction risk assessment method based on the Graph Attention Networks (GAT) model, which can consider the coupling relationship between projects. Firstly, a pre-construction risk assessment index system for power grid infrastructure projects was constructed to extract their nodal features. Secondly, we built the relationship network to extract two types of relationships between projects. Thirdly, we proposed the GAT model based on each type of relationship to predict the risk level of the project. Finally, the final evaluation results were obtained after combining model outputs based on each edge type. The empirical results showed that this method can quickly and accurately evaluate the risk level of power grid infrastructure projects before construction.