Using Bayesian network analysis to support centre of gravity analysis in military planning

Abstract

Centre of gravity (COG) analysis is an integral and cognitively demanding aspect of military operational planning. It involves identifying the enemy and friendly COG and subsequently determining the critical vulnerabilities that have to be degraded or negated to influence the COG of each side. This paper describes a modelling framework based on the causal relationships among the critical capabilities and requirements for an operation. The framework is subsequently used as a basis for the construction, population and analysis of Bayesian networks to support a rigorous and systematic approach to COG analysis. The importance of this work is that it uses existing planning process concepts to facilitate the construction of comprehensive models in which uncertainties and subjective judgements are clearly represented, thus enabling future re-use and traceability. The visual representation of the COG causal structure helps to clarify thinking and provides a way to record and impart this thinking. Moreover, it gives planners the capability to perform impact analysis, that is, to determine which actions are most likely to achieve a desirable end-state. The paper discusses the methodology, development and implementation of the COG Network Effects Tool (COGNET) suite for model population and model checking as well as impact analysis.

Introduction

Decision analysis typically makes use of operations research techniques such as probabilistic modelling, optimisation and game theory as a basis for analytical methods to evaluate and structure incomplete knowledge in order to reason about the implications of certain decisions. It is motivated by a need to understand “the current state of knowledge, its limitations and implications” (Morgan and Henrion, 1990) and provides a way to construct logical representations of the decision situation. In Barclay et al. (1997) a decision analysis methodology is presented, which is based on four elements: a set of initial courses of action; a set of possible consequences for each initial act; the value of each act in terms of money, utility or some other unit; and the likelihood that a particular act will result in a particular consequence. All four elements are considered during operational-level planning. The first two elements are an integral part of Course of Action (COA) development while the last two elements form the basis of systematic COA analysis.

The initial stage of an operational-level planning process typically includes some form of mission analysis. This involves identifying and analysing the superior commander’s intent in order to ensure that commanders and staff can determine which tasks are essential to achieve the operational objective. Correct assessment of the objective is deemed to be crucial to success at the operational level. The objective can be achieved by targeting the enemy’s centre of gravity (COG) through their vulnerabilities while protecting one’s own. In this thinking, the operational objective and the COG are inexorably linked. The COG, a key concept of operational art, can be thought of as a focal point that gives a force purpose and direction (see the article by Echevarria (2003) for an interesting discussion on the COG concept developed by Clausewitz and its subsequent interpretation).

Once the enemy COG has been determined, the planners must generate a suitable COA. Suitability refers to whether it meets the objectives as detailed in the mission analysis step. Since directly targeting the enemy COG is not usually feasible, a critical capability (CC) analysis is conducted at this stage of the planning process. A CC is a key element of a force that if destroyed, captured or neutralised will significantly undermine its COG. It does not necessarily have to be a military capability; it might be better described as a key factor contributing to the COG. Each CC might have a number of associated critical requirements (CR), which are essential for it to be fully functional. These requirements may be further decomposed into critical vulnerabilities (CV): elements that are potentially vulnerable. Each COA developed must target the enemy COG by exploiting the enemy’s CV in a sequence of actions. The planners must also identify CV from the enemy’s perspective, that is, related to the friendly COG. In summary, the planning team should identify the enemy’s COG and its associated CV, a number of approaches to undermine and neutralise it, as well as the own force’s COG and related CV.

A number of research projects are currently under way in DSTO’s Command & Control Division to develop decision-support tools with an underlying conceptual framework that complements the operational planning process but with a theoretical underpinning derived from decision analysis. All these tools feature interfaces that take into account the military user’s level of expertise and do not require a background in operations research. They include the COA Scheduling Tool (COAST) (Zhang et al., 2002), which provides a mathematical representation of a COA to enable quantitative analysis for sequencing and scheduling of tasks in an optimised COA; and COA Simulation (COA-Sim) (Matthews and Davies, 2003), a software agent-based wargaming tool to explore the feasibility, effectiveness and risk of an operational-level COA. The tools are being integrated using well-founded effects-based concepts to form an Integrated Modelling Environment (In-MODE). The research was motivated by a need for systematic and rigorous support for decision-making under uncertainty, which typifies the military operational environment.

This paper addresses one of the In-MODE research projects: the Centre of Gravity Network Effects Tool (COGNET), which uses causal probabilistic networks to represent the relationships among the CC and CR for a COG construct. COGNET models typically represent the functional decomposition of the COG to identify its influencing elements and to categorise them into a hierarchy: COG, CC and lower-level capabilities and requirements. This type of decomposition assumes that targeting a CR at the bottom of the functional hierarchy produces an effect on all related elements higher up. Using this model it is possible to investigate the effect that a set of actions has on the COG.

The paper starts by taking a look at probabilistic models, specifically Bayesian networks, and how they can be used to support operational planning. The concept of COG representation using Bayesian networks is then introduced and the characteristics of COGNET models are discussed. The following section describes the suite of tools that make up COGNET and how they can be used to construct, populate and analyse COG networks.