Multi-agent-based Solution Of 2d Water Flows On Gpus: A Versatile Platform For Interactive And Multi-perspective Flood Risk Research
Supervisor: Dr Georges Kesserwani
Joint Supervisor: Department
The provision of fast, accurate and flexible hydraulic forecasting models for flood resilience and risk management is a high priority research need (UK Flood and Coastal Erosion Research Strategy 2011-2030). Computer modelling enables real-time forecasts that can be used for emergency planning purposes such as evacuation and traffic management. This mandates hydraulic models capable of completing 12 to 24 hours flood forecasts in a few hours, or even minutes, to provide timely information to issue warning and support the respond in real-time. However, the wall-clock time required for an urban-resolution and/or large-scale flood simulation may far exceed the period of the flood event. The emergence of parallel computing on GPU (Graphical Processing Unit) has enabled run-time speed up of flood models. Still, the design of most computer flood models is a mesh-based finite element solution to a mathematical model (i.e. often the shallow water equations), which does not provide flexibility to further study with wider (none-mathematical) aspects of flooding (e.g. based on observation, random probability and social behaviourism).
This PhD aims to develop a fast and accurate real-time flood simulation tool, but within a holistic and flexible medium that allows integrating needs/skills of scientists and end-users from and across different sectors. The Flexible Large Scale Agent Modelling Environment for the GPU (FLAMEGPU) at the University of Sheffield provides parallel computational power within a multi-agents framework accessible without direct GPU programming. FLAME GPU will be used to (1) build a computer flood model by shifting from mesh-based passive modelling to a flexible Agent-Based interactive modelling on the parallel computing architecture of the GPU on a single computer; (2) Identify the extent (of spatial scale and memory size) to which direct forecast is possible? And (3) Integrate pedestrian agents in the model and its behaviour along with flood modelling. The PhD will be performed across Civil and Computer Science departmental to have access to the necessary skills and resources.
Suitable applicants are those with any Engineering, Mathematics Physics backgrounds, and computer science.
This project is NOT FUNDED, although Departmental/University scholarships are available for applicants who can demonstrate strong evidence of research potential.