Novel Applications of Computational Steering

Abstract

Large scale systems in science and engineering are often modelled by numerical simulation to monitor their behaviour and to provide the ability to forecast future state. With the increasing volume of data from observations in the natural and built environments, it is now possible to adjust parameters of such simulations so that they can track more accurately the actual state of the systems. The systems that are tracked by the simulations are named Dynamic Data Driven Application Systems (DDDAS) and the method that adjusts the simulations are already known as the Data Assimilation (DA) process. The DA utilized in DDDAS is often a compute-intensive and time-consuming process. Hence, existing DA applications can take hours and days to update the simulation. Although this updating speed is acceptable in investigations that are not subject to real-time constraints it is not acceptable in application that are, e.g. weather forecasting, monitoring of forest fires. Thus, a major challenge is to accelerate data assimilation process to prepare the simulation to keep pace with the development physical system being modelled. Computational steering has been utilized to optimize the exploration of parameter space in simulations. Based on the analysis of the simulation output, we hypothesize that adapting computational steering can also optimize data assimilation. This requires adaptation and automation of methods of utilizing computational steering so that data streams as well as human users can steer simulations. Consequently, we have developed a steering architecture to implement the steerable data assimilation process. To guarantee the updating frequency required by DDDAS, we introduce time management and computing resource management functions into the new computational steering architecture. To evaluate our hypothesis and steering architecture, we applied it to the problem of simulating the real-time behaviour of water distribution networks as an example of a DDDAS. Collaboration with the water utilities has provided real observation data for our experiments. We find that, by integrating computational steering with the data assimilation, the running time of the data assimilation method is dramatically decreased. Moreover, using steering results from the data assimilation, human users can also steer high-level parameters of the system. Furthermore, the time management and computing resource management functions are able to manage to control the running time of the steering process to enable real-time prediction of the behaviour of DDDAS.

Date of Award	31 Dec 2017
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Jonathan Shapiro (Supervisor)