International Journal

on Marine Navigation

and Safety of Sea Transportation

Volume 5

Number 3

September 2011

359

1 INTRODUCTION

Widely introduced on ships and land-based facilities,

navigational information systems are designed to as-

sist users in the decision-making process. Their main

task is collecting and presenting information neces-

sary for safe navigation. Currently available tech-

nologies give wider opportunities to assist naviga-

tors in interpretation of navigational situations and to

generate suggestions of possible solutions. This is

connected with the observed development of these

information systems in the direction of decision sup-

port systems. These systems are based on the use of

knowledge, which includes: the existing legal regu-

lations, procedures of conduct, principles of good

seamanship, navigational theories. The point is to

appropriately acquire that knowledge and create its

representation, which enables its efficient and effec-

tive use. The problem of decision support also ap-

plies to the interpretation of the navigational situa-

tion in accordance with the International Regulations

for Preventing Collisions at Sea (COLREGS). Due

to the fact that in certain areas local law may apply,

it is also important to take it into account. Assuming

that decision support also includes propositions of

solutions (e.g., manoeuvres), a knowledge base

should contain the principles of good seamanship.

The knowledge engineering is a branch of infor-

mation technology which deals with issues of

knowledge acquisition, representation and sharing.

The methods and tools of this new discipline open

way to the construction of systems facilitating the

interpretation of the sea route regulations and the

application of the principles of good seamanship.

The knowledge base built in this way can be used as

one of the expert system elements, which may be

part of a larger decision support system.

2 ASSUMPTION

There are many methods of knowledge representa-

tion that can be used in the navigational decision

support process. The most important are:

− decision trees, as a representation of possible

paths of decision-making depending on the exist-

ing and changing conditions;

− logical rules, presented in a simple or complex

form, contain the premises and conclusions re-

sulting from them;

− frames that are the base of object-oriented repre-

sentation of knowledge.

One of the main problems which appears during

the design of the knowledge base consists in choos-

ing the proper method of knowledge representation.

The decision to choose the proper ways of reasoning

is as important as the explanation method and future

expansion of knowledge. The selection of these

methods is strongly determined by the specificity of

the application field, therefore it is important to

identify assumptions for a knowledge base and fur-

ther for an expert system.

Knowledge Base in the Interpretation Process

of the Collision Regulations at Sea

P. Banas & M. Breitsprecher

Institute of Marine Technology, Maritime University of Szczecin, Szczecin, Poland

ABSTRACT: The article presents the problem of transforming knowledge contained in the provisions of the

International Regulations for Preventing Collisions at Sea, and the so called good seamanship in computer

applications. Some methods of knowledge representation in decision support in avoidance of collision situa-

tions are compared and examined. Acquisition, representation and sharing of knowledge are taken into con-

sideration from the viewpoint of supplementing the knowledge database and computational complexity.

360

The knowledge base should satisfy the following

requirements:

− ability to reproduce and verify the rules arising

from regulations contained in COLREGS;

− possibility of supplementing the knowledge with

special regulations issued by competent authority;

− opportunity to submit informal knowledge con-

tained in the principles of good seamanship;

− open database that allows easy expansion of

knowledge;

− simple presentation of the knowledge stored in

the database to make it understandable to people

who are not systems designers, and easy to verify.

The expert system using the above knowledge

base should satisfy the following requirements:

− generation of explicit answers resulting from the

knowledge contained in the database;

− interpretation of COLREGS regulations and spe-

cial rules;

− generation of proposals of actions to be taken

arising from the principles of good seamanship;

− explanation of the system response by quoting

relevant rules or principles.

An example of the implementation of COLREGS

is, designed at the Maritime University of Szczecin,

the knowledge base included in the Navigational

Decision Support System (NDSS) (Pietrzykowski et

al. 2009, Wołejsza 2005). The method of knowledge

representation used in this system has a form of de-

cision trees. They give a possibility of checking the

subsequent conditions and, on that basis, determin-

ing the response of the system. Their main disad-

vantage is the difficulty of extension, which may

cause considerable complexity of the system, and

thus the difficulties in the verification of correctness

and prolonged time to a response.

Knowledge representation based on logical rules

(rule-based knowledge base) is open to change, ena-

bles easy verification of accumulated knowledge and

makes it possible to offer explanations by simple

methods. Due to data security and the systems appli-

cation field, designers decided that the expert system

based on this knowledge base would not be a self-

learning system.

3 DECISION SUPPORT SYSTEMS

Knowledge bases are an integral part of decision

support systems. According to the theory of infor-

mation systems, it is possible to distinguish several

classes:

− transaction systems, whose main task is to collect

full information from trustworthy sources and

through simple models to allow the basic analysis

and compilation;

− management information systems, which aggre-

gate data from other systems, among them trans-

action systems; they use cross trend analysis us-

ing equation models, that operate on deterministic

data, but there may occur shortcomings and con-

tradictions;

− decision support systems, which have an extend-

ed data analysis mechanism for probabilistic, of-

ten contradictory, incomplete and incorrect; when

they assist the decision making process, they use

knowledge bases, and optimization and simula-

tion models;

− expert systems, built with the use of knowledge

and skills of people who are experts in the appro-

priate field, which are given as logic and heuristic

models;

− artificial intelligence systems, which use a wide

range of artificial intelligence methods to perform

modelling and analyses.

Because of the wide use of Decision Support Sys-

tems, in addition to the foregoing division, they can

be characterized by taking into consideration many

other factors and criteria. For example, the division

can be based on decision-making model or decision-

making process modelling (description, prescrip-

tion). The division by type of controlled system or

process comes down to the determination whether

the model used is deterministic, statistical or fuzzy.

The number of steps in decision-making process (in

one step or in n steps), the number decision-makers

involved in decision-making process (individual de-

cision or group decision) and the time factor (how

the time span is determined to take actions and deci-

sions) are further factors. The next factor character-

izing the decision support system is the manner it

works – static, when it stops just after giving the an-

swer or dynamic – system that at a given time dis-

cretization and at the occurrence of certain events

(inputs) operates continuously and appropriately

adapts to the existing conditions. Despite the divi-

sions and classifications, which in fact may reflect

many aspects, decision support systems, in general,

deal with data acquisition, convert information into

knowledge and generate answers (decisions) on the

basis of the knowledge they comprise.

This division of classes also shows the evolution

process of information systems from static to fully

dynamic, managing the models in decision support

systems (Pietrzykowski et al. 2007). It should also

be noted that expert systems currently being devel-

oped and artificial intelligence systems are ranked as

a subclass of decision support systems.

It is planned that the developed knowledge base

will be an element of the system considered as an

expert system. With its open formula, cooperation is

also possible with the systems that belong to other

classes.

361

4 EXPERT SYSTEMS

A typical expert system could be described as struc-

ture presented below:

− user interface – a module which is responsible for

interaction between the system and the user (nav-

igator). The module input allows user to ask ques-

tions and to specify additional information. The

module also gathers input data from other sys-

tems. The module output provides the user with

answers and explanations;

− knowledge base – data which are characterised by

special structure adjusted to store logical rules.

The rules are quickly searched for in accordance

with given criteria;

− inference mechanism – main part of the system

that is responsible for the process of solving the

problem. To do so, the mechanism uses

knowledge base, logical rules and additional in-

formation;

− explanation mechanism – part of the interface that

provides the explanation and gives legitimacy for

the system answer (decision) (Giarratano & Riley

2004, Jackson 1998).

The most important issues that shall be examined

during the process of creating knowledge base of the

expert system which offers COLREGS interpretation

and gives solutions for aiding navigational decisions

are:

− database structure definition;

− data acquisition.

Because the expert system with rule knowledge

base is to be considered, specific problems that

should be worked out are as follows:

− rules based on COLREGS and local laws or regu-

lations;

− inference mechanisms that are supposed to seek

proper rules and formulate conclusions;

− priorities mechanism that is responsible for build-

ing proper hierarchy of exact rules and regula-

tions;

− working out the inference mechanism based on

good seamanship procedures, that works simulta-

neously with the mechanism that considers

COLREGS rules and local regulations.

5 KNOWLEDGE ACQUISITION

The rules stored in the knowledge base are derived

from regulations provided in the COLREGS. The

regulations were taken under examination that led to

a set of logical rules. The connection between regu-

lations and rules was taken into consideration in

rules’ notation. It is essential during the explanation

process of the answer given by system. The opera-

tion of extracting data from special rules can be per-

formed in the same way.

In case of extracting any informal knowledge that

accounts for the principles of good seamanship, the

process is more complex. For example, Jones or

Cockroft diagrams which apply to restricted visibil-

ity situations can be treated as such source of

knowledge. The knowledge itself must be then

checked, verified and supplemented by experts -

navigators.

6 SOLUTION

The knowledge base and the system that are pro-

posed will contain a database divided into three parts

that apply to:

− rules from COLLISION REGULATIONS,

− rules that concern the principles of good seaman-

ship,

− additional rules that arise from special regula-

tions. It is important to bear in mind that local

rules have higher priority than general rules.

Therefore, it is possible that some conflicts will

occur and corresponding local and general rules

may be contradictory.

Ultimately, the proposed expert system is ex-

pected to work as stand-alone module or as be a part

of Navigational Decision Support System, devel-

oped at the Maritime University of Szczecin.

Figure 1. General architecture of navigational decision support

system on a seagoing vessel (Pietrzykowski & Uriasz 2009)

The diagram above shows arrangement of the

proposed system. It will replace the knowledge base

module and take over its function and tasks (Fig.1).

Due to the fact that the knowledge base will be

used for building the expert system and consequent-

ly it will operate in the decision support system, the

following structure of the expert system is taken into

consideration (Fig.2).

362

Figure 2. Structure of the proposed expert system.

The functions of the proposed expert system are:

− Data acquisition – the module that acquires data

required for the system to work, such as: position,

speed, navigational statuses, headings, relative

bearings, aspects, weather conditions, etc.;

− Data analysis – the module that analyses acquired

data, performs data standardization and calculates

formulas and parameters to be used in inference

mechanisms;

− COLREGS and special rules inference engine –

inference mechanism which performs the inter-

pretation of knowledge contained in the rules

stored in knowledge bases; on the basis of data

sent from ‘Data analysis’ module, optimal rules

are selected – the rules that fit the examined situa-

tion. Then, the selected rules are used in forward

chaining process and, consequently, the navigator

is informed if the vessel is a give-way or stand-on

one. In addition, the rules and special rules which

are applicable at the moment will be presented.

− Good seamanship inference engine – inference

mechanism running in parallel to the above men-

tioned one, using a knowledge base containing

rules reflecting the principles of good seaman-

ship; diagram of operation is very similar to the

previous one, but the output gives a suggestion of

conduct in a particular situation; the operation of

this module is independent from the previous one,

so it is possible to give suggestions for appropri-

ate conduct in spite of an absence of priority such

as manoeuvres resulting from the Cockcroft dia-

gram for restricted visibility;

− Knowledge bases – set of knowledge bases con-

taining rules resulting from the extraction of

knowledge from COLREGS, local regulations

and the rules of good seamanship, supplemented

with links to relevant regulations and explana-

tions;

− Presentation of results – this module prepares the

information obtained from inference mechanisms

for presentation and transfers them to the user in-

terface.

Data needed to carry out the inference are derived

from a host computer or directly from the navigation

systems of the vessel. As assumed, the proposed sys-

tem always includes all information available. If

some data are required and are not among those

transmitted, it may be necessary to refine data, tak-

ing place via the master system. Obtained data are

processed in order to adapt them to the internal rep-

resentation of knowledge present in the system. This

processing is primarily responsible for standardiza-

tion and for searching for contradictions (e.g., in-

compatibility of data obtained from different

sources, taking into account malfunctions, measure-

ment error or human factor). In the proposed system,

input information is obtained in the form of vector

data and it may contain contradictions that must be

eliminate. The processed data are the basis for the

inference process.

Inference mechanisms work with dedicated

knowledge bases, which contain rules extracted from

the COLREGS, local regulations (special rules) and

rules of good seamanship, a transcript of the

knowledge of expert navigators, supplemented by

other sources of non-codified rules for determining

the proper conduct in different situations. The meth-

od of writing the rules was adapted to the transferred

input values obtained from the data analysis module,

enabling the user to find them promptly.

For the storage of rules there is a dedicated data-

base with appropriately designed structure. Defined

in this structure are the tables for each set of rules

and additional data used in the process for request-

ing and managing the work of the whole expert sys-

tem. The first step in the process of inference is per-

formed by a query to the appropriate tables, which

results in a set of rules satisfying the conditions

compatible with the existing navigational situation.

Then, the provided rules of inference will be carried

forward, resulting in the generation of a conclusion,

passed to the module output.

Because there exist different sets of rules for the

COLREGS (and local regulations) and the principles

of good seamanship, two conclusions are obtained in

parallel: on the right way with an indication of ap-

plicable regulation and rules, and the proposed pro-

cedure in the situation.

Figure 3. Functional diagram of data acquisition.

363

Figure 4. Functional diagram of the inference process.

The system operation is based on the following

algorithm shown in Figures 3, 4:

1 Get input from the user interface and navigation

equipment;

1 Adapt to the requirements of the input search

rules;

2 Find rules matching the analysed situation in the

respective knowledge bases;

3 If rules were found in the database with local reg-

ulations, they must be taken into account in deci-

sion making;

4 If rules were found in COLREGS, determine their

hierarchy against the local regulations;

5 Decide on the basis of the rules contained in the

databases of local and COLREGS;

6 If emergency arises resulting from the absence of

rules and regulations of local COLREGS, acquire

again data (clarify) (go to point 1). If the data

cannot be refined, send information about the

lack of data required to make a decision;

7 If the base of the principles of good seamanship

contains rules for the situation, determine sugges-

tions resulting from these rules;

8 Show the results of the expert system work.

Marine navigation equipment and systems on

board are the source of input data for running the

expert system based on the knowledge base. Where

an ambiguous situation is identified, that may result

from the inability to establish certain facts, a request

will be made to supplement data, for example by a

dialogue between system and navigator.

Rules in the knowledge base should be supple-

mented with additional information, such as the de-

gree of validity of the a rules e.g. rules with the low-

est degree of validity are used only to carry out the

inference process, rules of higher degree will also

appear on the screen as a warning, and a rule having

the highest degree will additionally activate a sound

alarm.

If the system has an additional knowledge base

that contains rules based on local laws, it is possible

to automatically take it into account when making a

decision. The moment of inclusion or exclusion de-

pends on the geographical position of the vessel ob-

tained from navigational devices (e.g. GPS).

7 EXAMPLES

Below are examples of the proposed expert system

functioning. The system receives selected input data

and provides a response in the form of conclusions

and explanations. The system response and risk of

collision which is determined by the decision sup-

port system provides the basis for working out the

manoeuvre according to COLREGS. The answer is

preceded by an analysis of the rules contained in the

knowledge base. Examples 1 to 4 are shown on fig-

ure 5.

1 Input data: restricted visibility = NO; Distance =

5Nm; Own priority (S1) = 6 (power-driven ves-

sel); Other vessel’s priority (S2) = 6; Relative

bearing = 005° PS - Port Side; Aspect = 150° SS -

Starboard Side; Own speed (V1) > Other vessel’s

speed (V2)

Conclusion: Give a way

Explanation: conclusion based on rule No. 13 of

COLREGS – Overtaking.

2 Input data: restricted visibility = NO; Distance =

3Nm; Own priority = 6; Other vessel’s priority =

6; Relative bearing = 100° SS; Aspect = 040° PS;

Own speed < Other vessel’s speed

Conclusion: Give a way

Explanation: conclusion based on rule No. 15 of

COLREGS – Crossing situations.

3 Input data: restricted visibility = NO; Distance =

3Nm; Own priority = 6; Other vessel’s priority =

6; Relative bearing = 003° SS; Aspect = 004° PS;

Own speed ≈ Other vessel’s speed

Conclusion: Head on vessels, Alter your curse to

starboard

Explanation: conclusion based on rule No. 14 of

COLREGS – Head-on situation.

4 Input data: restricted visibility = YES; Distance =

2Nm; Own priority = 6; Other vessel's priority =

6; Relative bearing = 010° SS; Aspect = 010° PS;

Own speed ≈ Other vessel's speed

Conclusion: all actions should be taken to avoid

collision – reduce your speed.

Explanation: conclusion based on rule No. 19 of

COLREGS (all actions should be taken to avoid

collision).

5 Input data: restricted visibility = NO; Distance =

10Nm; Own priority = 6; Other vessel’s priority =

6; Own speed ≈ Other vessel’s speed

Conclusion: none

Explanation: The distance is considered as safe.

364

Figure 5. Examples of navigational situations.

Explanatory notes and assumptions applied to the

above mentioned examples

− 8Nm is a distance assumed to be safe by the sys-

tem and no actions need to be taken in naviga-

tional situation;

− Vessel’s priorities gradation is adopted from sim-

plified “privileged hierarchy” (Rymarz 1995).

8 SUMMARY

The construction of decision support systems can

significantly expand the capabilities of navigational

decision support systems. Their important element is

the knowledge base, which may be a separate struc-

ture or part of expert systems. The effectiveness of

decision support systems largely depends on the cor-

rectness of the knowledge base, its structure and

mode of action.

The following assumptions and proposals of im-

plementation have been presented:

1 To use an expert system as a sub module of deci-

sion support system;

2 To prepare a knowledge base based on rules;

3 To divide a database to 3 parts containing respec-

tively COLREGS rules, special rules and princi-

ples of good seamanship rules to improve effi-

ciency and to allow the knowledge to be updated.

REFERENCES

COLREG 1972. Convention on the International Regulations

for Preventing Collisions at Sea. International Maritime

Organization

Giarratano J.C. & Riley G.D 2004. Expert Systems, Principles

and Programming

Jackson P. 1998. Introduction to Expert Systems

Pietrzykowski Z., Chomski J., Magaj J. & Bąk A. 2007. Aims

and tasks of the navigational support system on a sea going

vessel, In J. Mikulski (ed.) Advanced in Transport Systems

Telematics 2, Publisher Faculty of Transport, Silesian Uni-

versity: 251-259. Katowice

Pietrzykowski Z., Magaj J. & Chomski J. 2009. A navigational

decision support system for sea-going ships, Measurement

Automation and Monitoring

Pietrzykowski Z. & Uriasz J. 2009.Knowledge representation

in a ship’s navigational decision support system. In Adam

Weintrit (ed.), Marine Navigation and Safety of Sea Trans-

portation:45-5., Gdynia

Rymarz W. 1995. Podręcznik Międzynarodowego Prawa Drogi

Morskiej. Gdynia

Wołejsza P. 2005. An Algorithm of an Anti-collision Manoeu-

vre. Międzynarodowa Konferencja Naukowo-Techniczna

Inżynieria Ruchu Morskiego. Świnoujście