Bayesian hierarchical methods for modeling electrical grid component failures

Abstract

This paper presents a probabilistic modeling framework for component failures in electric power grids. Failure probabilities for grid components are often estimated using parametric models informed by past observations of failure. This work formulates a Bayesian hierarchical framework designed to integrate data and domain expertise to understand the failure properties of a regional power system, where variability in the expected performance of individual components gives rise to failure processes that are heterogeneous and uncertain. We use Bayesian methods to fit failure models to failure data generated in simulation. We test our algorithm by evaluating differences between the data-generating model, our Bayesian hierarchical model, and maximum likelihood parameter estimates. We evaluate how well each model can approximate the failure properties of individual components, and of the system overall. Finally, we define an upgrade policy for achieving targeted reductions in risk exposure, and compare the magnitude of upgrades recommended by each model.