Agent-Based Modelling, the Next 20 Years: Dynamic Data Assimilation


Nick Malleson, Jon Ward, Andy Evans

Schools of Geography & Mathematics, University of Leeds

nickmalleson.co.uk

surf.leeds.ac.uk

These slides: http://surf.leeds.ac.uk/presentations.html

Abstract

Agent-based modelling is proving itself to be an extremely valuable approach to understanding and modelling social systems. In particular, the treatment of individuals directly - rather than through aggregate statistical rules - places it in an ideal position to model dynamical systems that depend on the actions and interactions of numerous individuals. Part of the success of the methodology can be attributed to its interdisciplinarity; researchers from disciplines ranging from sociology and psychology to computer science and physics have contributed to its development.

However, agent-based modelling has yet to incorporate some key ideas from fields such as meteorology, particularly with respect to improving models dynamically in response to streams of new data. Part of the explanation of this absence can be attributed to data availability - only recently have data sources begun to emerge that have the spatio-temporal resolution and volume to warrant the incorporation of methods similar to those used in the environmental sciences.

This paper begins to address this gap by experimenting with an Ensemble Kalman Filter (EnKF) as a means of dynamically assimilating data into an agent-based model of urban dynamics. The paper will demonstrate that an agent-based model coupled to an EnKF can be a reliable means to dynamically improve predictive models in response to streaming data such as that arising from social media or through devices that count the number of people passing certain points. Importantly, these data do not need to track individuals, which reduces the ethical risks associated with such invasive surveillance. Ultimately the model presented here will be used to create estimates of the size of the population at risk to phenomena such as crime victimisation or to ill health as a result of exposure to air-borne pollutants.

Overview

Background - agent-based models and making future predictions

Barriers to ABMs as predictive tools: difficult to calibrate

1. Computationally Expensive

2. Data hungry

3. Divergent

Dynamic Data Assimilation (DDA) & ABM - difficulties and outlook

Agent-Based Modelling (ABM)

Autonomous, interacting 'agents'

Model phenomena from the 'bottom-up'

Advantages:

Modelling complexity, non-linearity, emergence

Natural description of a system

Bridge between verbal theories and mathematical models

History of the evolution of the system

ABM Example - Burglary

Malleson, N., A. Heppenstall, L. See, A. Evans (2013) Using an agent-based crime simulation to predict the effects of urban regeneration on individual household burglary risk. Environment and Planning B: Planning and Design 40 405-426. [DOI: 10.1068/b38057]

 

 

 

 

 

 

 

 

 

ABM Problems

1. Computationally Expensive

Not amenable to machine-led calibration

2. Data hungry

Need fine-grained information about individual actions and behaviours

3. Divergent

Usually models represent complex systems

Projections / forecasts quickly diverge from reality

Who cares?

ABM growing in importance

Smart cities

Pedestrian simulations

Evacuation

Traffic

. . .

The most appropriate methodology for simulating complex (social) systems

(Potentially) drive urban planning decisions in a smart cities context

1. Computationally Expensive

Very large parameter space to explore

Models are usually highly non-linear

And very difficult to explore it

Lots of agents * lots of iterations * lots of model runs

But not so much of a problem as it once was

Simulating Urban Flows (surf) prototype

github.com/nickmalleson/surf

Source: Improbable Spatial OS http://improbable.io/learn-more
Example of different modelling scales
Pattern oriented modelling. Source: Grim et al. (2005)

2. Data Hungry

Data required at every stage

Particularly for calibration and validation

But high-quality data are hard to come by

—Many sources are too sparse, low spatial/temporal resolution

Censuses focus on attributes rather than behaviour and occur infrequently

Also need data from numerous scales (e.g. pattern oriented modelling: Grimm et al. 2005)

Understanding social behaviour

How to estimate leisure times / locations? Where to socialise?

Comparing census daytime populations to aggregate mobile phone estimates
Evaluating aggregate mobile phone population accuracy

Rise of Big Data

Recent explosion in data volume.

'Datafication'

Streams of data

Great potential for calibrating ABM

Example: Mobile Communications

'Big Data'

Example: Social Media

Example activity spaces derived from twitter messages
Malleson, N and M. Birkin. (2014) New Insights into Individual Activity Spaces using Crowd-Sourced Big Data. Paper presented at the 2014 BigData conference, Stanford, CA, USA, 27-31 May. [paper (pdf)][slides (html)].
Traces of movement from a smart-phone app

'Big Data'

Example: Geo-Apps

Smart-phone apps that capture movement / location are becoming ubiquitous

Great potential for understanding (some) urban dynamics

3. Divergence

Complex systems

One-shot calibration

Nonlinear models predict near future well, but diverge over time.

The process of calibration
Typical model development process

3. Divergence

Drawback with the 'typical' model development process

Waterfall-style approach is common

Calibrate until fitness is reasonable, then make predictions

But we can do better:

Better computers

More (streaming) data

Methodological gap

Dynamic Data Assimilation

Used in meteorology and hydrology to constrain models closer to reality.

We can:

Restart model with new (up to date) data

Adjust model parameters to better match reality

Dynamic Data Assimilation

How?

Particle filters

Indoor footfall (Rai and Hu, 2013.; Wang and Hu, 2015)

Kalman Filter

Air traffic (Chen et al., 2012)

Ensemble Kalman Filter

Pedestrian footfall (Ward et al., 2016)

Sequential Monte Carlo (SMC)

Wildfire (Hu, 2011; Mandel et al., 2012)

Ensemble Kalman Filter (EnKF)

Outline

Ward, J., A. Evans, N. Malleson (2016) Dynamic calibration of agent-based models using data assimilation. Royal Society Open Science. 3:150703. (open access). [DOI: 10.1098/rsos.150703]

Ensemble Kalman Filter (EnKF)

Maths

Broad literature, but generally tied to mathematical models (e.g. differential equations and linear functions)

Not applicable to ABM (inherently non-linear)

Working with a mathematician to do the hard work!

In all its glory: Ward et al., (2016)

Advantages

Similar to Kalman Filter (best in class)

But better for nonlinear systems

Conclusion and Outlook

Potential for ABM to be better at making future predictions

Computing power

Big Data

Data assimilation

Adapt methods used commonly in other fields

Potential for ABM + DDA + Smart Cities ...

Chen, Haiyan, Jiandong Wang, and Lirong Feng. 2012. “Research on the Dynamic Data-Driven Application System Architecture for Flight Delay Prediction.” Journal of Software 7 (2). doi:10.4304/jsw.7.2.263-268.

Grimm, Volker, Eloy Revilla, Uta Berger, Florian Jeltsch, Wolf M. Mooij, Steven F. Railsback, Hans-Hermann Thulke, Jacob Weiner, Thorsten Wiegand, and Donald L. DeAngelis. (2005) Pattern-Oriented Modeling of Agent-Based Complex Systems: Lessons from Ecology. Science 310(5750): 987–91. doi:10.1126/science.1116681.

Hu, Xiaolin. 2011. “Dynamic Data Driven Simulation.” SCS M&S Magazine 1: 16–22.

Mandel, Jan, Jonathan D. Beezley, Adam K. Kochanski, Volodymyr Y. Kondratenko, and Minjeong Kim. 2012. “Assimilation of Perimeter Data and Coupling with Fuel Moisture in a Wildland Fire–Atmosphere DDDAS.” Procedia Computer Science 9: 1100–1109. doi:10.1016/j.procs.2012.04.119.

Rai, S., and X. Hu. 2013. “Behavior Pattern Detection for Data Assimilation in Agent-Based Simulation of Smart Environments.” In 2013 IEEE/WIC/ACM International Joint Conferences on Web Intelligence (WI) and Intelligent Agent Technologies (IAT), 2:171–78. doi:10.1109/WI-IAT.2013.106.

Wang, M. and X. Hu (2015). Data assimilation in agent based simulation of smart environments using particle filters. Simulation Modelling Practice and Theory 56, 36–54.

Ward, Jonathan A., Andrew J. Evans, and Nicolas S. Malleson. 2016. “Dynamic Calibration of Agent-Based Models Using Data Assimilation.” Open Science 3 (4). doi:10.1098/rsos.150703.

Thank you

Agent-Based Modelling, the Next 20 Years: Dynamic Data Assimilation


Nick Malleson, Jon Ward, Andy Evans

Schools of Geography & Mathematics, University of Leeds

nickmalleson.co.uk

surf.leeds.ac.uk

These slides: http://surf.leeds.ac.uk/presentations.html