Towards Graph Machine Learning for Smart Grid Knowledge Graphs in Industrial Scenarios

Knowledge Graphs (KGs) demonstrated promising application perspective in different scenarios, especially when combined with Graph Machine Learning (GML) techniques able to interpret and infer over facts. Given the natural network structures of Smart Grid equipment and the exponential growth of electric power data, Smart Grid Knowledge Graphs (SGKGs) provides unprecedented opportunities to manage massive power resources and provide intelligent applications. However, a single representation of the SGKGs is never sufficient to properly exploit GML techniques that leverage different aspects of the KG for various objectives. In this work, we provide a methodology to extract various significant views of the SGKG by iteratively applying a series of transformation to the description of the power network in the IEC CIM standard. Our implementation is based on a declarative approach to guarantee easier portability, and we deploy the transformations as a stateless microservice, facilitating modular integration with the rest of the Smart Grid Semantic Platform. Experimental evaluation on two real power distribution networks demonstrates the efficacy of our approach in highlighting important topological information, without discarding precious additional knowledge present in the SGKG.