Causality modelling of simulated temporally indexed events to construct a combat narrative
Files
(Published version)
Date
2021
Authors
Tiong, J.K.R.
Chiera, B.A.
Editors
Advisors
Journal Title
Journal ISSN
Volume Title
Type:
Conference paper
Citation
Proceedings of the International Congress on Modelling and Simulation, MODSIM, 2021, pp.897-903
Statement of Responsibility
Conference Name
24th International Congress on Modelling and Simulation, MODSIM 2021 (5 Dec 2021 - 10 Dec 2021 : Sydney, Australia)
Abstract
Combat simulation studies are commonly used to support decision-making on the basis of competing outputs, simulated under comparable conditions, for a selection of input design points. Such simulators have long been in use to support analysis in force design, operational requirements, mission area analysis and force-on-force analysis, necessitating high-resolution, closed-form and stochastic simulation capabilities. Each simulation typically runs as a black-box, making it difficult to discern the intermediate stochastically-varying series of events that form part of the combat narrative, detailing the sequence of events that led from the simulator inputs to the observed outputs. Although the simulator runs in black-box mode, internal simulator data on stochastically-varying temporally indexed events over the combat spatial region are also captured. These events record an ordered series of events that, although stochastically seeded, reflect the fundamental logic of cause and effect. Understanding these event series through appropriate statistical modelling has the potential to provide key insights into the combat mission narrative and support decision-making around combat missions. In this paper we present a first step towards constructing a combat mission narrative to support decision-making via a case study of internal combat simulation event series and their associated outputs. Statistical modelling of the internal data is challenged by the underlying simulator mechanisms - the use of Common Random Numbers (CRN) to reduce the variability in simulation output. CRN simulators are fraught with difficulties as they violate key statistical assumptions by design, requiring alternative, robust methods of analysis. We adopt the use of the statistically robust Event Coincidence Analysis (ECA) to capture causality between events by providing a framework for quantifying the strength, directionality and time lag between two event series. The use of ECA is novel in this area of application; ECA has been recently adopted in the literature, predominantly in the areas of ecology, environment and health and is relatively under-explored in Defence, with current areas of application including armed conflict and hate-speech triggered terrorism. An attractive feature of ECA is that it allows for significance testing of causality between two series, based on stochastic point processes with a prescribed inter-event time distribution and other higher-order properties, thus providing a differentiation between coincidence and causal events.
Specifically, ECA considers two types of causal behaviour - precursor and trigger - with the former describing a series of events that typically occur before a secondary event takes place (mediated cause-and-effect) whereas the latter captures the concept of direct cause-and-effect between two events. Extensions of ECA include aggregation and conditionalisation; the former providing an integrated measure for coincidences that occur between several pairs of event series subject to some meaningful grouping mechanism and the latter allowing for the flexibility of interlinking multiple causal event series, to allow for conditioning of events on specific situations. The R library coincalc was used to implement ECA as part of constructing a combat narrative. The case study provided four event series of interest - Movement, Detections, Shots and Kills - yielding combat narratives around the progression of Movement leading to Detection, being Shot and a Kill. Suggested combat narratives arising from the analyses conducted herein were: (i) standing can act as either a precursor or trigger to being identified or recognised while detection is a mediating, not direct, trigger for being shot. These results were supported by two different ECA methodologies; and (ii) for those simulations resulting in combat mission failure, precursory behaviour led to the failure of the overall mission over a relatively short time window, rather than a single, or series of, direct triggers.
School/Discipline
Dissertation Note
Provenance
Description
Access Status
Rights
Copyright 2021 Proceedings of the International Congress on Modelling and Simulation, MODSIM