Aggregation of stochastic models

Hjorth, Poul (2007) Aggregation of stochastic models. European Study Group with Industry > 56th ESGI [Bath 3/4/2006 - 7/4/2006].

Full text available as:

PDF - Requires Adobe Acrobat Reader or other PDF viewer.

Abstract/Summary

Dstl examines battle simulations based on stochastic evolution codes. One such code, known as SIMBAT (SIMple BATtle program) models the evolution of the battle as a sequence of turns in which the two sets of combating units move in a landscape. The units have objectives and act accordingly, they can acquire enemy units if lines-of-sight in the landscape permit this, and they can fire upon and disable enemy units with a certain probability. Based on the setting of a large number of parameters, and also on the outcome of pseudo-random decisions and engagements made in the course of the action, a final outcome of the battle is achieved. Figure 1 shows a SIMBAT screenshot.

Figure 1: Screenshot from a SIMBAT battle. The setting is that of a ‘standard’ battle (see full report).

The problem posed to the Study Group is the following: Is it possible to analyse and subsequently calculate a battle in terms of smaller subunits which can then be aggregated into the whole in a systematic fashion? This could potentially speed up the processing of a large number of simulations.

Item Type:	Study Group Report
Study Group:	European Study Group with Industry > 56th ESGI [Bath 3/4/2006 - 7/4/2006]
Company Name:	Dstl
Industrial Sector:	Aerospace and defence
Additional Contributors:	Williams, Rhys and Collander-Brown, Simon and Gould, Tim and Zyskin, Maxim
ID Code:	106
Deposited By:	Richard Booth
Deposited On:	19 June 2007

Problem Statement

Dstl simulate battles by means of stochastic evolution codes. These involve detailed simulation of many units in the battlefield. The Study Group was asked to investigate methods of ‘aggregation’ that would simulate a battle in simpler terms by aggregating the units together, so that simulations could be accelerated.

The Study Group proposed partitioning the battlefield into zones and treating the numbers of combatants in those zones as continuous variables obeying ordinary differential equations, possibly with stochastic terms. The parameters in those equations, and the stochastic terms, would need to be determined by running small, unit-to-unit, combats in the full simulator, which can be done much more quickly than simulating the whole battle.

Archive Staff Only: edit this record