b'Smart ContingencyCombining fundamental physics-based relationships in the electrical grid Analysis Neural Networkinfrastructure with artificial intelligence-based analyses enables power utility for in-depth Power Gridproviders to explore and discover system vulnerabilities before problems occur.VulnerabilityAnalyses C ontingency analyses are crucial to prevent detrimental power outages by enabling utilities to discover vulnerabilities within their grid. An n-k contingency is subject to k number of component failures caused by severe weather conditions, software failure (e.g., 2003 Northeastern blackout), or cyberattacks (e.g., 2019 denial-of-service attack on the Western transmission grid). However, power utilities are typically limited to n-1 system-wide contingency analysis due to the combinatorial nature of the problemthe computational TOTAL APPROVED AMOUNT:cost required to sweep through all possible power flow simulations increases $120,000 over 1 year exponentially as a function of k. This research investigated the potential of a smart PROJECT NUMBER:contingency analysis neural network (SCANN) framework based on machine 20A1047-038 learning to reduce the computational expense normally required to calculate all possible n-k contingencies by half. SCANN is used to discover higher order PRINCIPAL INVESTIGATOR:contingencies up to n-3, after which the prediction error becomes greater than the Bjorn Vaagensmith acceptable tolerance. Additionally, SCANN can compute results 147 times faster than CO-INVESTIGATOR: the Newton-Raphson solver used in conventional power grid analyses. Based on this Sam Yang, Florida State University learning with SCANN, which uses residual neural networks, a new research focus into the resiliency of the grid using Bayesian neural networks can be explored.TALENT PIPELINE:Deepika Patra, student at Arizona State University.PRESENTATION:Yang, S., B. Vaagensmith, and D. Patra, Power grid contingency analysis with machine learning: A brief survey and prospects, IEEE Resilience Week (RWS) (2020).SCANN uses data-driven encoding to learn relationships between power grid components.124'