Data assimilation for agent-based models

The aim of this project is to develop a computational and mathematical framework for data assimilation using agent-based models (ABMs), which integrates data emerging from smart cities (e.g. traffic counters, social media activity, environment sensors, etc.) into large-scale urban simulations in real time.

Socio-economic systems, where individuals' behaviour is dependent on the actions and interactions of the system's constituent parts, can be modelled in an intuitive and natural way using ABMs. To date, the primary use of such models has been to explain macroscopic phenomena in terms of individual agent behaviour, and explore the effects of different scenarios. However, the need for considerable data on which to parameterise and validate such models has generally held back agent-based socio-economic modelling. The recent emergence of streamed 'big' datasets provide a significant opportunity for ABM calibration and validation in real-time, and even for the use of ABMs as predictive tools.

Data assimilation techniques have been developed and used extensively in weather forecasting in order to combine forecasts from computational fluid dynamics models with live data streams. This project will investigate the different data assimilation challenges that ABMs pose and develop new techniques to solve them. It will be necessary to develop computational Bayesian inference methods that can be used with cloud-based ensemble simulations of ABMs. The focus of the project is on methodological development, so there is some flexibility in terms of the specific socio-economic models that could be implemented.