393
1 INTRODUCTION
Theprimarytasksofanavigationofficerincludethe
collectionandinterpretationofinformationaswellas
the forecast of future occurrences through
conscientious watch. Based on this perspective,
apprenticeofficersareobservedtolacktheexperience
requiredtoconductatimely andrealisticawareness
ofasituation,
whichoftenresultsinsomeinaccurate
interpretations or erratic predictions. Sometimes,
despitetheaccurateinterpretationofthesituation,the
action to avoid collision is conducted late,
endangering the safety of the ship or situation. An
effective resolution to ensure safe navigation is
simulation‐based training that has been essentially
designed
to equip apprentice officers with the
required set of skills and experience. Lee (2018)
confirmedthatsimulationofthenavigationscenarios
will help the apprentice officers to improve their
awarenessandabilities.
Karlsson(2011)denotedtheimportanceofbriefing
and debriefing for ensuring effective training and
presented the evaluation items to
students during
simulation and debriefing. Park (2016) provided
standardized evaluation items with respect to the
simulation operators. These studies identified the
factors of focus during the training as evaluation
items;however,theydidnotprovidetheimportance
andquantitativeguidelinesforsuchitems.
A previously conducted study suggested a
guideline
forensuringsafenavigationbyattempting
to define the optimal possible speed for sea voyage
legs(Rutkowski,2016).Thisstudyfurthersuggesteda
Analyzing the Factors Affecting the Safe Maritime
Navigation for Training Apprentice Officers
M.K.Lee
OceanScienceandTechnologySchoolofKoreaMaritimeandOceanUniversity,Busan,SouthKorea
S.W.Park
KoreaMaritimeInstitute,Busan,SouthKorea
Y.S.Park&M.J.Park
KoreaMaritimeandOceanUniversity,Busan,S.Korea
E.K.Lee
SafeTechResearch,Inc.,Daejeon,S.Korea
ABSTRACT:Oneoftheprimaryfactorsthataffectthesafemaritimenavigationistheinsufficientexperience
and skill of an apprentice officer, which may be improved using simulation‐based training by ensuring
operationalefficiency.Thisstudyaimstodetermineappropriatefactorsforachievingeffectiveandintensive
simulation
‐basedtrainingofapprenticeofficersandpresenttheguidelinesforsuchatrainingscheme.Initially,
amarinetrafficriskmodel,whichinterpretsandaccuratelymeasurestheriskofcollisionwithothervessels,is
analyzed to derive the most influential factors in safe navigation. Subsequently, simulation experiments are
conductedbyapplying
machinelearningtoverifytherequiredsafenavigationfactorsforeffectivelytraining
theapprenticeofficers.Asaresultoftheaboveanalysis,itwasconfirmedthatthefactoraffectingsafemaritime
navigation was the distance from other vessels. Finally, the differences between these distances in the
simulations are analyzed for both
the apprentice officers and the experienced officers, and the guidelines
correspondingtoboththesecasesarepresented.Thisstudyhasthelimitationbecauseofthedifferencebetween
theshipmaneuversimulationandtheactualshipnavigation.Thiscanberesolvedbasedontheresultsofthis
study,incombinationwith
theactualnavigationdata.
http://www.transnav.eu
the International Journal
on Marine Navigation
and Safety of Sea Transportation
Volume 13
Number 2
June 2019
DOI:10.12716/1001.13.02.17
394
method of preparing a table for ensuring safe
minimum distances during a navigational watch
(Rymarz,2007),similartotheonethatwassuggested
basedonthesafetyconsciousnessoftheKoreanship
operators(Parketal.,2010),andproposedabasicVTS
guideline with the ship’s closest point of
approach/time
to the closest point of approach
(CPA/TCPA), collision risk, control frequency, and
minimum safety distance through VHF
communication analysis (Park et al., 2017).
Regardless,thesestudieswereconductedwithrespect
tothegeneralnavigation officers and did not reflect
thecharacteristicbehaviorsoftheapprenticeofficers.
Further, this study intends
to derive the items
intended for ensuring the extensive and effective
trainingofapprenticeofficersthroughananalysisof
the maritime traffic risk assessment model and the
simulation experiment. Additionally, the guidelines
arepresented quantitativelyafter respective analysis
ofthesimulationresults.
2 EXTRACTIONOFTHEEXTENSIVETRAINING
FACTORS
2.1 Analysis
basedonmarinetrafficriskmodel
2.1.1 PARKmodel
The potential assessment of risk factors (PARK)
modelwasusedto evaluate the risk involvedinthe
vessel operators navigating the coastal waters of
Korea (Ngyuen, 2014). Here, the risk factors of the
shipoperatorwereclassifiedasinternalfactors,such
astheshiplength,tonnage,shiptype,andrankofthe
officer, and external factors, such as the crossing
direction, ship speed, and separation distance
between ships. In this model, the risk value can be
calculatedasfollows:
12
/
5.081905
0.002517
0.004930ꞏ 0.430710ꞏ
pf f
ffff f
fiop d
Risk value T T L
WC LP LC
SH S S D
(1)
where,
p
T Own ship type factor ;
f
T Own ship ton factor Ton ;
f
W Own ship width factor m ;
1f
C ship operator s career factor
;
f
L license factor ;
f
P position factor ;
L LOA oftheownship(m);
2f
C crossing situation factor ;
f
S side approaching factor ;
i
o
H inner / outer harbor factor ;
p
S own ship speed factor kt ;
d
S speed difference between ships kt ;and
DdistanceNM .
Only the factors related to the other ship were
identifiedbecausetheinformationabouttheownship
remained constant. Figure 1 denotes the degree of
thesefactorswithrespecttothefinalrisk.
Figure1.RiskfactorbyPARKmodel
ThePARKmodelestimatedthattherangeofrisk
is1–7,thesensitivityofthedistancefactoris2.15355,
the factor of the speed difference between ships is
0.12325, the speed factor is 0.1777098, the inner and
outerharborfactoris 0.062305, thesideapproaching
factoris0.118905,crossingsituationfactor
is0.158458,
and the length overall (LOA) of the other ship is
1.0068. Among these, the distance was 2.15355,
confirming that the distance factor affect on
approximately30.7%ofthetotalrisk.
2.1.2 ESmodel(ESS)
The environmental stress (ES) model is based on
thedifficultyassociatedwiththeoperationoftheship
accompanied by the load imposed on the operator
andattemptstoquantifythenatural,terrain,facility,
and traffic conditions surrounding the operator
(Inoue et al., 1998). This model is subjected to two
types of stress, including stress with respect to the
environment of operation and stress with respect to
the traffic environment. However, this study was
limitedonlybytheanalysisofthetrafficenvironment
stressrelevanttotheshiptraffic.Thedeterminantsof
thetrafficenvironmentstressincludedthedistanceto
oppnentshipandtheaveragelengthoftheownship
andthe opponent ship. In thismodel,the risk value
canbedeterminedasfollows:
Risk value α ( / / ) β
α/ αR
α 0.00192 Lm
RVV Lm
RLm
(2)
Crossingfactorwithothership
starboard crossing β 0.65 ln Lm 2.07
port crossing β 0.65 ln Lm 2.35
Head on β 0.65 ln Lm 2.07
Overtaken β 0.65 ln Lm 0.85
The ES model is based on the difficulty of ship
operationaccompanied by therestrictionof the load
imposed on the operator when the surrounding
conditionssurroundingtheoperator,suchasnatural
conditions,terrainconditions,facilityconditions,and
trafficconditions,itwasamodelthattriedtoquantify
(Inoue et al., 1998). The ES model has stress on the
environment of operation, and stress on the traffic
395
environment.However,thisstudylimitedlyanalyzed
thetrafficenvironmentstressrelatedtothetrafficof
theship.Thedeterminantsofthetrafficenvironment
stress are the distance to theopponent ship and the
averagelengthoftheownshipandothership.
Figure 2 illustratesthe degree of the factors with
respecttothefinalriskbasedontheESmodel.
Figure2.RiskfactorbyESmodel(ESS)
The range of risk was 0–6, the sensitivity of
distancewas6,andtheaveragelengthoftheownand
opponent ships was 1.354532. Among these, the
distance factor was 6, confirming that the distance
factoraffecton100%ofthetotalrisk.
2.2 Analysisbasedonmachinelearning
2.2.1 Classification
modelconstructionbasedonRFC
algorithm
Random forest (RFC) algorithm is a type of
ensemblelearning method used in classification and
regressionanalysisinmachinelearning.Itoperatesby
outputting class (classification) or average predicted
value (regression analysis) from a plurality of
decision trees constructed in the training process
(Park,
2017). In this study, classification model was
constructedbyusingRFCmodelthroughanalysisof
navigationpatternandtheimportanceofthefeatures
was assessed. Simulation experiments were
conductedtocollectthedatatoconstructthemodel.
1 Simulationoverview
TheKanmon Straitwas selectedasthetestwater
bodybecause
ofitshighlevelofdifficulty(Shinet
al., 2017). Additionally, because of the highest
trafficvolumefrom8:00amto 9:00 am,this time
zone was selected, where the corresponding AIS
datawasreceivedandthecorrespondingscenario
wascreated(Hiroakietal.,2010;Lee,2018).
Figure3.Testwaters(KanmonStrait)
Thesubjectsareconsistedoftwogroups,oneisthe
apprentice officer group who is the 4th grade
student at the Korea Maritime and Ocean
Universitywithoneyearofonboardtraining,and
the other is experienced officer group. Figure 4
illustratesthecareerofexperiencedofficergroup.
Figure4.Rankandexperienceofthetestofficers
7 teams were formed and each team including a
captain who are experienced in Kanmon Strait.
The familiarization were conducted in different
parts of the test waters and repeated more than
twicetominimizetheinfluenceofshipmaneuver
simulationfamiliarity.Thetestsweresubsequently
conducted.
2 ConstructionofClassificationmodel
For analyzing the navigation patterns, the
followingdata were collected:shipspeed at each
time zone, vessel size considered to be the most
dangerous, distance between the vessels, DCPA,
TCPA, PARK model risk value, and encounter
situation. The random forest algorithm was
employed for constructing a classification model
bylearning
using80%ofthedataandperforming
modeltestingusingtheremaining20%ofthedata.
Currently, the average score obtained using the
testdatawas0.87.
3 Evaluationoftheclassificationmodelandfeature
importance
Table 1 presents the evaluation results for the
constructedmodel.
Table1.Evaluationindexoftheconstructedmodel
_______________________________________________
Precision Recall F1‐score
_______________________________________________
Apprenticeofficer 0.88 0.94 0.91
Experiencedofficer 0.83 0.67 0.74
Average0.86 0.86 0.86
_______________________________________________
In Table 1, precision indicates the ratio of actual
True that the machine learning model evaluated as
True,whereasrecallindicatestheratioofthecorrect
answer value of the model to the correct answer
value.
F1‐score denotes the harmonic mean value
obtained using precision and recall as appropriate
(Park,
2017).Notethattheconstructedmodeldenoted
396
reliability with average values of 0.86 for precision,
recall,andF1‐score.
Subsequently, a receiver operator characteristic
(ROC) graph was created for the model as other
evaluationmethods.TheROCgraphwasobtainedby
visualizing the horizontal axis to denote a false
positive rate and the vertical axis to denote
a true
positive rate. This model uses values of the area
under curve (AUC) to represent the accuracy as a
singlenumber,andtheROC–AUCvalueistheareaof
theROCgraph.Figure5displaystheROCgraphfor
thecreatedmodel.
Figure5.ROCgraphforconstructedmodel
FromFig.5,AUCexhibitsavalueof0.94,whichis
ahighscore.Thus,theevaluationmethodsconfirmed
the reliability of the model. The importance of each
featurewasanalyzedusingthismodeltodistinguish
between the apprentice officers and the experienced
ones.
Figure6.Featureimportance
Figure 6 displays the importance of theanalyzed
features by distinguishing the apprentice or
experienced officers in order of distance with other
vessel>TCPA>PARKmodelrisk>speed>DCPA>
secondaryshiplength>crossingfactor.Thedistance
denotedthehighestimportanceof0.243.
2.3 Sub‐conclusion
In order to acquire the factors that can enhance the
safe navigation skill by intensively training the
apprentice officers, marine traffic risk model was
analyzed and the navigation pattern was analyzed
usingmachinelearning.
First,ThePARKandESmodelswereanalyzedto
evaluatethemarinetrafficriskandderive
thefactors
that should be included in the extensive training of
apprenticeofficerstoacquirethenecessarynavigation
skills. There was a significant variation between the
riskfactorsforboththemodels,exceptintermofthe
separation distance, which were observed to be the
factors that contributed most to safety
risks.
Particularly,the distance reflecteda 100%maximum
riskpossibilityintheESmodel,andthePARKmodel
determined that the risk contributed by the distance
couldconstituteupto30.7%ofthetotalrisk.
Secondly,thenavigationpatternswereanalyzedto
identifythemostimportantfeaturesthatcanbe
used
todistinguishtheexperiencebetweentheapprentice
and experienced officers. As a result, the distance
from other vessels was identified as the most
importantfactorwiththeimportanceof0.243.
Thereforeitwasconfirmedthatthatitisimportant
fortheapprenticeofficerstosailortoacquiretheskill
to
maintain a constant distance from other ships for
safemaritimenavigation.
3 COMPARATIVEANALYSISRESULTOF
EXTENSIVETRAININGFACTORS
3.1 T‐testfordistance
Toverifywhethertherewasastatisticaldifferencein
themeanofthedistancebetweentheapprenticeand
theexperiencedofficers,aT‐testwasconducted
based
ontheencounterrelationpresentedinFig.7.
Figure7.Encounterdegrees
The null hypothesis is that the average distance
betweentheapprenticeandexperiencedofficerswas
the same; the alternative hypothesis is that the
average distance between the apprentice and
experiencedofficerswasdifferent.Table2reflectsthe
resultsoftheT‐test.
397
Table2.ResultsoftheT‐test
_______________________________________________
N M SDT(p)
_______________________________________________
Head‐onApprentice 235 764.22 171.58−3.949
Experience 228 820.77 135.03 (.00)***
FineCrossingApprentice 558 744.81 180.80 2.408
(St’bd) Experience 222 707.24 202.53 (.017)**
BroadCross. Apprentice 504 651.09 201.28 4.474
(St’bd) Experience 155 566.21 222.94 (.00)***
Converging Apprentice 518 609.27 228.54 3.050
Cross.(St’bd)Experience 154 544.15 245.91 (.002)**
Overtaking Apprentice 1622 670.41 224.85 5.283
Experience 953 621.79 226.52 (.00)***
FineApprentice 634 733.77 178.69−1.566
Cross.(port) Experience 493 750.71 182.22 (.118)
Broad Apprentice 731 681.41 214.43 2.440
Cross.(port) Experience 271 640.90 240.11 (.015)**
onverging Apprentice 590 536.96 202.04 1.625
Cross.(port) Experience 270 510.57 229.17 (.105)
_______________________________________________
p*<0.1,p**<0.05,p***<0.01
Exceptforfineandconvergingcrossingattheport,
the null hypothesis was rejected; hence, the
alternative hypothesis was adopted. In short, the
average distance navigated by majority of the
apprentice and experienced officers was different.
Particularly,asconfirmedbytheaveragevalues,the
experienced officers navigate longer distances than
thatnavigatedbytheapprenticeofficers.
3.2 Density
The separation distance between the ships was
confirmedby plottingevery 10s using thedirection
and relative distance. Figures 8 and 9 display the
scatterplotsoftherelativedistances.
Figure11.Shipplottingincaseoftheapprenticeofficer
Figure10. Relative distance scatter plot in case of the
experiencedofficer
Basedonthesefigures,thejuniorofficerseemedto
maintain a space of approximately 150 m without a
ship,nearly200minfrontandapproximately750m
on both the sides. The density of the ship was
provided by a mesh of 50 m to verify the ship
distribution,as
depictedinFigures10and11.
Figure9.Meshofshipsincaseoftheapprenticeofficer
Figure8.Meshofshipsincaseoftheexperiencedofficer
Note that the ship distribution in case of the
apprentice officer safe separation distances between
thefrontandrearoftheshipwereuncertain.Onthe
contrary, the safe separation distances are observed
whentheexperiencedofficersnavigate,wherea400m
398
separation in the forward direction and a 200m
separation in the backward direction were
maintained.
3.3 Minimumandaverageofrelativedistanceeach
bearing
The separation distances were validated using
relative bearing, where the average and minimum
relative distances were denoted graphically by
dividingthebearingaroundtheshipsinto
10°units.
Figures 12 and 13 shows the average and minimum
distancebybearing.
Figure13.Averagedistancebybearing
Figure12.Minimumdistancebybearing
Theaveragedistanceobtainedbyrelativebearing
was 651 m for the junior officer and 598 m for the
experiencedofficer,indicating a smalldistance yield
for the latter. In contrast, the averages of the
minimumdistanceswere135mforthejuniorofficer
and 171 m for the experienced officer,
indicating a
largedistanceforthelatter.Especially,theminimum
distance yield for theexperienced officers was large
forboththefrontandrearoftheship.Thisconfirms
that the total distance in case of the experienced
officer was shorter than that in case of the junior
officers.However,
thesuitableclearancedistancewas
maintainedinthefrontandrearoftheshipincaseof
theexperiencedofficers.
4 CONCLUSION
Thisstudyintendedtoderivethefactorsthatshould
be included for the extensive training of apprentice
officerstoensurethat they acquire the necessaryset
ofskillsand
experienceand,subsequently,tosuggest
the corresponding guidelines. The factors were
selectedthroughtheanalysisoftheriskfactorsduring
navigationusingthemarinetrafficriskmodelandthe
machine learning algorithm. Further, simulations
were conducted for both groups of junior and
experienced officers to derive the static value of
the
separation distance of ships, which exhibits the
greatest influence among the risk factors. The
conclusionscanbegeneralizedasfollows:
1 The PARK and ES models were employed and
analyzedforselectingthe most influential factors
while evaluating the navigation risk factors that
should be included for the extensive training
of
apprenticeofficers.IntheESmodel,theseparation
distancebetweenthe ships exhibiteda maximum
influence of 100%. In the PARK model, the
separation distance attributed to the maximum
riskwas30.7%.
2 Based on the results of navigation simulation in
Kanmon Strait by apprentice and experienced
officers,a
classificationmodelwascreatedbyRFC
algorithmusingmachinelearning.Here,thefactor
with the greatest influence on classification was
the distance factor at 0.243, which denoted the
highest importance among all the features. Thus,
thedistancefromothervesselswasconsideredto
beanextensivetrainingitem.
3 The results
of the simulations conducted with
respecttothejuniorandexperiencedofficersthat
aimedtoderiveanappropriateseparationdistance
yielded statistically significant values, except for
thedirectionof247.5°to330°.
4 Afteranalyzingthedensityoftheothershipsand
their relative distances from the own ships
according
tothedirections,shortaveragedistances
were observed for the experienced officers even
though the separation distances in the front and
rearwerelong.Specifically,aseparation distance
of 400 m was maintained 350°–010° in the front
andthatof200mwasmaintained170°–190°inthe
rear.
As a result,
a safe separation distance of
approximately 600 m was confirmed including the
frontandtherear.Thus,theminimumsafeseparation
distanceshouldbemaintained600matKanmonStrait
and this should be educated inexperienced junior
officersasaguideline.
In the future studies, the simulations could be
extended to
different test waters to achieve
appropriate safe separation distance in various
scenarioswhenthenavigationisbeingperformedby
junior officers and comparative analysis with the
actualnavigationdatashouldbeperformedtoprove
the appropriateness of the minimum safe separation
distance.
399
REFERENCES
E.W. Rymarz. (2007): The Determination of a Minimum
Critical Distance for Avoiding Action by a Stand‐on
Vessel as Permitted by Rule 17a) ii, TransNav, the
International Journal on Marine Navigation and Safety
ofSeaTransportation,Vol.1,No.1,63‐68
G. Rutkowski. (2016): Determining Ship’s Safe Speed and
BestPossibleSpeed
forSeaVoyageLegs,TransNav,the
International Journal on Marine Navigation and Safety
ofSeaTransportation,Vol.10,No.3,425‐430
HiroakiS.,MiraiO.,HideoU.,MasatoshiS.(2010):Marine
Traffic Analysis of Three Major Bay – Ship Handling
DifficultyofShipEquippedwithAIS,JapanInstituteof
Navigation,Vol.123,
No.3,13‐19
Inoue, K., Kubono, M., Miyasaka, M., Hara, D. (1998):
ModelingofMariners’PerceptionofSafetywhenBeing
Faced with Imminent Danger (In Japanese), Journal of
JapanInstituteofNavigation,235‐245
KarlssonT.(2011):Theimportanceofstructuredbriefings&
debriefingsforobjective evaluation ofARPA simulator
training,
ChalmersUniversity,Ph.D.
LeeMyoungki(2018):AStudyontheEvaluationofRADAR
/ ARPA Simulation Training Results, Korea maritime
andOceanUniversitygraduateschool,MasterThesis.
Nguyen,X.T.(2014):ABasicStudyontheDevelopmentof
Real Time Supporting System (RTSS) for VTS Officers,
KoreamaritimeandOcean
Universitygraduateschool.
Ph.D.
Park Sang‐won, Park Young‐soo, Park Jin‐soo (2017): A
Study on Basic VTS Guideline based on Shipʹs
Operatorʹs Consciousness, TransNav, the International
Journal on Marine Navigation and Safety of Sea
Transportation,Vol.11,No.4,597‐603
ParkYoung‐Soo(2016):AStudyonthe
Standardizationof
EducationModulesforARPA/RadarSimulation,Journal
oftheKoreanSocietyofMarineEnvironment&Safety,
Vol.22,No.6,631‐638
ParkYoung‐Soo,JeongJae‐Yong,KimJong‐Sung(2010):A
Study on the Minimum Safety Distance between
Navigation Vessels based on Vessel Operator’s Safety
Consciousness,Journalofthe
KoreanSocietyofMarine
Environment&Safety,Vol.16,No.4,401‐406
ParkHae‐sun(2017):Introductiontomachinelearningwith
python,HanbitMedia,Inc.115‐121,335‐360
Shin Dae‐Woon, Park Young‐Soo, Kim Dae‐Hae (2017): A
StudyontheeffectARPA/RADARSimulationTraining,
JournaloftheKoreanSociety
ofMarineEnvironment&
Safety,Vol.23,No.3,294‐300
