Technology Evolution-oriented Graph Representation Learning for Hidden Innovation Relationship Discovery in Power Grid Networks

Abstract This study proposes a Power Grid Innovation Graph Representation Learning (PGI-GRL) framework for technology evolution-oriented hidden innovation relationship discovery in power grid networks. Existing approaches mainly rely on bibliometric analysis, text mining, or conventional graph learning methods, which are insufficient for modeling the evolutionary characteristics of heterogeneous innovation relationships and discovering latent technological connections. Scientific publications and patent documents are integrated to construct a heterogeneous innovation graph, where multiple innovation entities and relationship types are modeled to characterize different technological evolution patterns. Based on this graph, semantic features, structural dependencies, and heterogeneous relationship information are jointly learned through a technology evolution-oriented graph representation learning mechanism, enabling the framework to capture the evolutionary significance of different innovation interactions. A relationship prediction module is further developed to identify potential technological connections and latent innovation pathways. Experiments conducted on real-world power grid innovation data evaluate the proposed framework through link prediction, temporal relationship prediction, and hidden innovation discovery tasks. The results demonstrate that the proposed framework consistently outperforms representative graph embedding and graph neural network methods, showing superior capability in modeling complex innovation evolution patterns and discovering potential technology combinations. The proposed framework provides an effective data-driven approach for understanding technology evolution and supporting innovation relationship discovery in power grid networks.