Power line modems (PLMs) act as communication devices inside a power line network (PLN). However, they can be exploited also as active sensors to monitor the status of the electric power distribution grid. Indeed, power line communication (PLC) signals carry information about the topological structure of the network, internal electrical phenomena, the surrounding environment and possible anomalies in the grid. An accurate and efficient identification of the types of anomaly through direct sensing measurements can enable grid operators to both prevent malfunctions and effectively intervene when faults occur. In this paper, we present how to use supervised machine learning (ML) techniques to extract anomalies information from high frequency measurement of electrical quantities, namely the line impedance, the reflection coefficient and the channel transfer function, in the PLC signal band. Simulation results confirm the potentiality of the neural network method, outperforming existing model-based approaches in the field without any hyperparameter tuning.
