Searching for risk in large complex spaces

Abstract

ASHiCS (Automating the Search for Hazards in Complex Systems) uses evolutionary search on air traffic control simulations to find scenario configurations that generate high risk for a given air sector. Weighted heuristics are able to focus on specific events, flight paths or aircraft so that the search can effectively target incidents of interest. We describe how work on the characterization of our solution space suggests that destructive mutation operators perform badly in sensitive, high dimensional spaces. Finally, our work raises some issues about using collective risk assessment to discover significant safety events and whether the results are useful to safety analysts.