PMU placement for fault line location using neural additive models—A global XAI technique

In recent years, machine learning (ML) techniques have shown impressive results in many power system applications, including line fault location using phasor measurement units (PMU). Despite this success, such techniques may be hard to use in real life scenarios, if their results are not well understood by power system experts. In this light, this work proposes to use a global Explainable Artificial Intelligence (XAI) model for the problem of line fault location. Using such a model allows the user to examine which bus measurements are used for classifying different line faults, increasing user understanding and trust. Another benefit of this technique is that it provides a method for PMU placement in cases where the system is partially observable. The XAI method used is a Neural Additive Model (NAM), which provides a global explanation. Using this model and a novel evaluation technique where norm weights in the model are used to determine which bus measurements are most impactful, the most important measurements are determined, allowing for better PMU placement. This technique is compared to existing methods and matches benchmark performance while being computationally cheaper.