Topology-Based Machine Learning: Predicting Power Line Communication Quality in Smart Grids

Abstract

Smart Grids (SG) envision the exchange of both power and data, enabling system and customers to generate and transfer energy in a more efficient and balanced way. Among the relevant communication technologies, we find Power-Line Communications (PLC), which allow for data transmission on the electrical cables used for power delivery. Despite the hostile medium, PLC offer reliability and data rates to support exchange of control traffic, smart metering and sensor network applications. The distribution portion of the power delivery network, which we focus on in this work, is topologically complex, which makes channel prediction complicated. We show how random realistic topologies can be generated and then used to train a Machine Learning (ML) algorithm to infer PLC link quality (based on channel response) exploiting solely topology descriptors. We eventually show how precisely the communication quality can be inferred from the SG topology through ML. In doing so, we also discuss how the ML approach offers the common ground between top-down and bottom-up approaches for network characterization and how it enables smart decision making in the SG.

Keywords

• Bottom-up channel modeling
• machine learning
• network performance
• power line communication
• smart grid
• theoretical evaluation