Large-scale failures in stochastic flow networks are disastrous events that occur in many different applications, e.g. communication networks, transportation systems and power grids. This is particularly troublesome when the failure sizes are heavy-tailed: when extremely large failure sizes are substantially more likely to occur than one would infer from more conventional statistical laws. We model and analyze heavy-tailed phenomena in stochastic networks with a particular focus on cascading failures phenomena. We study how heavy-tailed input elements drive the heavy-tailed nature of the failure sizes, while the internal flow dynamics have a nontrivial but minor contribution. We exploit this fundamental insight to develop meaningful mitigation strategies that significantly decrease failure sizes and probabilities using learning algorithms.
|Supervisors||Fiona Sloothaak (TU/e) and Maria Vlasiou (TU/e)|
|PhD Student||Agnieszka Janicka|
|Location||Eindhoven University of Technology (TU/e)|