Dealing with uncertainty in agent-based models for short-term predictions

Abstract

Agent-based models (ABM) are gaining traction as one of the most powerful modelling tools within the social sciences. They are particularly suited to simulating complex systems. Despite many methodological advances within ABM, one of the major drawbacks is their inability to incorporate real-time data to make accurate short-term predictions. This paper presents an approach that allows ABMs to be dynamically optimised. Through a combination of parameter calibration and data assimilation (DA), the accuracy of model-based predictions using ABM in real time is increased. We use the exemplar of a bus route system to explore these methods. The bus route ABMs developed in this research are examples of ABMs that can be dynamically optimised by a combination of parameter calibration and DA. The proposed model and framework can also be used in an passenger information system, or in an Intelligent Transport Systems to provide forecasts of bus locations and arrival times.A PREPRINT -AUGUST 23, 2019 varies over the route and with at off-peak to peak periods. There are methods to incorporate streaming data into models, such as data assimilation (DA) routines [26,45]. Broadly, DA refers to a suite of techniques that allow observational data to be incorporated into models [45] to provide an optimal estimate of the evolving state of the system. Performing DA increases the probability of having an accurate representation of the current state of the system, thereby reducing the uncertainty of future predictions. This is a technique that has been widely applied in fields such as meteorology, hydrology and oceanography [21].There are, however, two methodological challenges that must be overcome to apply DA in ABM. First, DA methods are often intrinsic to their underlying models which are typically systems of partial differential equations with functions linearised mathematically. Hence DA methods typically rely on linearising the underlying model [16]. One of the most appealing aspects of agent-based models is that they are inherently non-linear, so it is not clear whether the assumptions of traditional DA methods will hold. Second, it is still unknown how much uncertainty DA can effectively deal with when implemented within ABM. Assimilation of real-time data into ABMs has only been attempted a few times and these examples are limited by their simplicity [28,46,47]. This paper is part of a wider programme of work 2 that is focused on developing DA methods to be readily used in ABM. This paper focuses on one particular model that aims to make predictions of bus locations in real time. Bus route operation has been chosen due to its inherent uncertainties -for example a model will need to account for uncertain factors affecting how buses travel on the roads [22] -but also for its tractability -there are many fewer interactions than present in, say, a model of a crowd. We also focus on one particular DA algorithm -the Particle Filter (PF). This method is chosen due to its ability to incorporate data into non-linear models such as ABM...

Exploring city digital twins as policy tools: A task-based approach to generating synthetic data on urban mobility

“…Agent-based modeling of human mobility in the city requires finegrained micro-level data for input; however, this type of data is often not available due to numerous reasons including privacy concerns (Bektas and Schumann, 2019). Another drawback of the agent-based modeling approaches is that they are not properly predictive (GeoTwin, 2020) and unable to incorporate real-time data flows for short-term predictions (Kieu et al, 2020).…”

“…The second part discusses the limitations of the current generation of CDTs. Concerns related to data privacy, availability, and their applicability for predictive simulations are discussed (Bektas and Schumann, 2019;Kieu et al, 2020) alongside the role that synthetic data can play in addressing these concerns (Nikolenko, 2019). The final section of the article proposes an alternative task-based approach to urban mobility data generation.…”

Healthcare and education networks interaction as an indicator of social services stability following natural disasters

“…Furthermore, it offers sufficient level of flexibility to accommodate different levels of system complexity in which features such as aggregation of agents, agent sub-components, and different levels of descriptions for agents can be represented 27 . However, these models have several limitations, including the uncertainty associated with their expected results 28 .…”

Improving Model Predictions—Integration of Real-Time Sensor Data into a Running Simulation of an Agent-Based Model

“…Saltelli [17] expands on this thought by identifying the potential of datasets that emerge from sensor networks to increase model accuracy by allowing for better calibration. Yet, Clay et al [18] and Kieu et al [19] state that it is currently not possible to use ABMs for real-time simulation due to the absence of established mechanisms for dynamically incorporating real-time data.…”