Harnessing Explainable Artificial Intelligence to Model the Causes of Unplanned Power Outages in Renewable Integrated Electrical Grids

Several developing countries not only suffer from electricity shortages due to low electrification rates, but also experience frequent power cuts and rolling black-outs, such as load shedding. These chronic grid-scale power outages are unplanned, and their causes are unknown, and have not been studied in detail in the current literature. Furthermore, their impact on economic growth and quality of life is well known to be detrimental. This study presents an innovative, explainable artificial intelligence grid outage model (XAI-GOM) that integrates Exponential Gaussian Process Regression with SHAP-based explanations to predict unplanned grid-scale load shedding outages. It provides data-driven and explanation-based policy recommendations on how electrical utilities can mitigate future power outages in the grid. When applied to a case study of a national grid, XAI-GOM's near-diagonal distribution reveals its strength in capturing complex nonlinear relationships by learning from 31 features, such as renewable generation and import/export flows. The results demonstrate that unplanned outages in coal-based power plants, combined with insufficient nuclear generation, remain the most significant contributors to elevated load shedding power outages. Based on the XAI-GOM, we can recommend enhanced asset management with predictive analytics to detect early signs of deterioration in thermal generation, preventing breakdowns that cause power outages. We can also make an XAI-GOM-informed recommendation to policymakers to expand solar, wind, and concentrated solar power with strong grid integration to mitigate variability and protect against unplanned thermal outages.