Interpretability of the ML-Based State of Risk Predictions for the Electric Grid Forced Outages

The electric grid forced outages (further referred to as “outages”) are increasingly common due to the growing complexity of the electric grid, wear and tear of the components, and the impact of inclement weather. The recent development of Machine Learning (ML) approaches for predicting the outage State of Risk (SoR) allows the time needed to implement mitigation measures that can reduce the risk and, consequently, impacts associated with the outages. This paper establishes a baseline for the outage SoR prediction using data from historical outage logs and relevant weather parameters associated with an actual power system in Texas, USA. The baseline is used to demonstrate how various design requirements for the outage SoR prediction models can impact the interpretability of the results. The paper’s findings emphasize the implications of different interpretability outcomes that affect the effectiveness of the mitigation options.