fire response test bench based on nfidnfidcanada.ca/wp-content/uploads/2018/03/final... ·...
TRANSCRIPT
Adriano O. Solis, Ali Asgary, Jenaro Nosedal‐Sánchez and Beatrice Zaccaro
March 1, 2018
Developing a fire response test bench based on NFID
Table of Contents EXECUTIVE SUMMARY 2 PROJECT OBJECTIVE 3 NFID: INITIAL ANALYSIS AND QUALITY ASSESSMENT 3
DATA AVAILABILITY AND CONSISTENCY 3
FIRES IN RELATION TO OTHER EMERGENCIES 6 INCIDENT DATASET: CURRENT CASE STUDY 12 INCIDENT GENERATION ENGINE 20 RESPONSE SIMULATION MODEL 23
MODEL DESCRIPTION 23
SIMULATION RESULTS (FUNCTIONALITY VALIDATION OF THE MODEL) 32 CONCLUSIONS AND FURTHER RESEARCH 38
EXTENSION AND IMPROVEMENT OF THE DEVELOPED MODELS 39 ACKNOWLEDGMENT 39 REFERENCES 39 AUTHOR BIOGRAPHICAL INFORMATION 40 LIST OF TABLES AND FIGURES 41
2
Executive Summary
Based on assumptions of completeness and consistency of fire incident data reported in theNational Fire InformationDatabase (NFID) for the 11‐year period from 2005 through 2015, theresearch team had proposed to develop, by leveraging relevant information in the NFID, asimulation engine that would provide fire departments across Canada with a tool for fireprevention,riskanalysis,preparedness,training,andresponsemanagement.
Our initialevaluationof thevariousNFIDdata fields,however,showedveryseriousgapsboth intermsofmissingdatavalues(blanks)aswellasapparent inconsistencies inthedataasreported.Moreimportantly,ourreviewofannualreportsofatleastfourcities(threeinOntarioandoneinAlberta)showedthatfireincidentsconstituteverysmallpercentagesofoverallincidentstheyhavereportedrespondingto.ThishasrendereddevelopmentofthesimulationmodelusingNFIDdata,aspreviouslyenvisioned,unattainable.WerequestedtheassistanceoftheVaughanFireandRescueService(VFRS),whoseofficialsagreedto provide relevant information in their 2009‐2016 database, subject to a non‐disclosureagreement between YorkUniversity and the City of Vaughan.Wewere provided full 2009‐2016incidentdatasets(includingrespondingunits,civiliancasualties,andfirefightercasualties).We proceeded to develop amodelling and simulation (M&S) framework involving two separatesimulation models: (i) an Incident Generation Engine, a discrete event simulation model usingcoloredPetrinets,whichcreates a listof incidentsbasedonempiricaldistributionsover timeofemergencyincidentsandtheirkeyattributes,and(ii)aResponseSimulationModel,anagent‐basedsimulation model which uses as inputs the list of incidents generated by the first model. Bothsimulationmodelshavebeentestedandpreliminaryresultsarereportedhere.Whileverification,validation and accreditation (VV&A) have constantly been conducted in the development andapplicationofthetwomodels,bothmodelsaremaystillbefurtherrefinedandimproved.1
1 This is anupdated versionof our researchproject report on ‘Developing a fire response simulation testbenchbasedonNFID’. TheproposalassubmittedtotheCanadianAssociationofFireChiefs inDecember2016hadspecifieda12‐monthprojecttimetable.Delaysinprojectapproval/contractingresultedinaccessto the NFID datasets by research team members commencing only on April 12, 2017, even as projectcompletion and report submission deadline remained specified as December 31, 2017.We submitted apreliminarydraftonthatdate,whileindicatingsubmissionofanupdatedreportbyMarch1,2018.
3
Project Objective
This research project was initiated to develop a simulation engine leveraging the National FireInformationDatabase(NFID)inordertoprovidefiredepartmentsacrossCanadawithadatadriventoolforevidence‐basedplanningandresponsetofireincidents,therebyhelpingthemcreatesafercommunities.
FinancialsupportwasprovidedintheformofagrantfromtheCanadianAssociationofFireChiefs(CAFC).
NFID: Initial Analysis and Quality Assessment
DATAAVAILABILITYANDCONSISTENCY
TheNFIDwasmadeavailabletotheresearchteaminitiallyonApril12,2017– intwomaindatafiles, one with information about incidents and the other containing information about victims(civiliansandfirefighters)forthe11yearsfrom2005through2015.Theincidentdatasetincluded128fields(columns)with467,929reportedincidents(rows),whilethevictimdatasetincluded30fields (columns)with15,326 reported cases (rows). In addition, anNFIDDataDictionary andanNFIDUserGuidewereprovided.
After initial analysis,we reported to the CAFCResearchGrantsAdministrator onApril 28, 2017aboutanumberofissues/concernswiththedatasets.Amongothers,wehadfoundthefollowing:
Multiple incidents with the same Incident ID (INCDNTID). In one very extreme case, forinstance,142incidentsreportedforSaskatchewanhadtheexactsameIncidentID.
Clarificationwas requiredwith respect todefinitionsof certain fields, including apparent
inconsistenciesinvaluesforcertainfields.Forexample,anentryof8intheBuildingHeight(HEIGHT) fieldmaybe interpretedeither in termsof abuildingwith8 storeysor as ‘Notapplicable(vehicle,outsidearea,etc.)’.
WewerereferredtoStatisticsCanada’sProjectLead,CanadianCentreforJusticeStatistics,Social,HealthandLabourStatistics,who– inaconferencecallwiththeresearchteamonMay4,2017–providedinsightsandclarificationonsomeofthefieldsinquestion.
On July 14, 2017, updated versions of the incident and victimdatasetswere released, aswell asslightlymodifiedversionsoftheNFIDDataDictionaryandtheNFIDUserGuide.The‘new’incidentdatasetcontainsthesamenumberofincidents,buthas136fields(columns).Notably,thefirstfieldinthisupdateddatasetistheLinkingID(LINK_ID)whichsequentiallynumberstheincidentsfrom1through467,929.
The current incident dataset accordingly contains 136 fields, which correspond to specificattributesthatshouldberecorded/reportedforeachincident,alldefined/explained,alongwiththe
4
codingvalues(i.e.,individualcategories,counts,magnitudes,etc.)intheNFIDDataDictionaryandtheNFIDUserGuide.
However,ourmacro‐analysisofboththeoriginalandupdatedversionsoftheincidentandvictimdatasets, we detected a very significant amount of missing values (blanks) that would haveconstitutedcriticalinputsforourmodellingandsimulation.
OurfirstfindingwasthattheNFIDdoesnotcontaininformationforallCanadianprovinces.Dataforonlysixprovinces(otherthantheCanadianArmedForcesasaseventhjurisdiction)arereported.Moreover,theNFIDdoesnotreport11yearsofincidentsforallthesesevenjurisdictions:
Only2005‐2014dataareavailableforOntario;datafor2015aremissing.
DataforSaskatchewancoveronlytheyears2012‐2015;datafor2005‐2011aremissing.
Intermsofthetemporaloccurrenceofincidents,thedataspreadsheetcontainsthefollowingfields:YEAR,MONTH,DATE,DAY, andTIME. In exploring data availability in theNFIDwith respect tothesefields,wefoundthefollowing:
YEAR:availablefor100%ofthelistedincidents.
MONTH and DATE: available for 95% of the listed incidents; no data reported for NewBrunswickandforCanadianArmedForces.
TIME: available for only 44 % of the reported incidents; not available for the entireincidentsrecordfromOntario.
Thefollowingfieldspertaintothelocationofeachfireincident:INCIDLOC(IncidentLocation),CSD(Census Subdivision Code),CSD_NAME (Census SubdivisionName),CMACA (CensusMetropolitanArea/CensusAgglomeration),CMA_NAME(CensusMetropolitan/AgglomerationAreaName).
INCIDLOC:availablefor94.4%ofthelistedincidents.However,responsesarenotprovidedinastandardizedform,whichimpairstheiruseforareliablespatialanalysis.
CSDandCSD_NAME:availablefor85.4%ofthelistedincidents.InthecaseofSaskatchewanonly0.2%ofthelistedincidentsreporttheCSD.
CMACAandCMA_NAME:availableforonly70.4%ofthereportedincidents.
Averyrelevantdatafieldforpurposesofmodellingandsimulationoffiredepartmentoperationalperformance are the times to respond to incidents. In the NFID, the response time in minutes(RESPONSE),referringtothetimebetweenreceiptofthealarm/callbytheFiredepartmenttothearrivalofthefirstresponders(i.e.,ResponseTimeofFirstVehicleatthesceneofthe incident), isoneofthefields.Unfortunately,thisinformationisquitescarceintheNFID.
RESPONSE:availableonlyforthejurisdictionofAlberta(13.2%ofthereportedincidents).
5
Infact,amoredetailedreviewoftheRESPONSEdataforAlbertaindicatesmanyinconsistenciesinthecoding.31.4%ofthevaluesarehigherthan60minutes(duetothecountofsuchvalues,thosecannot reallybe consideredaspotential statistical outliers).Moreover,30.5%of all the reportedvaluesare“999”,whichisclearlyproblematic.Afterremovingtheapparentlywronglycodedvalues,thedatagoesdowntoonly68.1%ofrecordedresponses,whichtranslates into9%oftheoverallincidentsreportedintheNFID.
Other relevant fields have to do with resources dedicated to the response, among others:CREWSIZE,NUMBEROFENGINES,NUMBEROFAERIALS,NUMBEROFTANKERS.
CREWSIZE:available for50.4%of thereported incidents,but, in fact, isonlyreportedforOntario.Oftheresponsesrecorded,45.2%arecodedas“0”(whichbearsnomeaning),andthe remaining values range from 1 to 251 ‐ which all clearly suggest the necessity of averificationofmanyofthereportedvalues.
NUMBEROFENGINES:entriesavailableforlessthan1%ofthelistedincidents.
NUMBEROFAERIALS:entriesavailableforlessthan1%ofthelistedincidents.
NUMBEROFTANKERS:entriesavailableforlessthan1%ofthelistedincidents.
Since the locations of the incidents are not reported in a consistently usefulway, an alternativerelevant piece of information for our research project is the reported distance from the firstresponderlocationtotheincidentlocation.ThisinformationisprovidedbytheDISTANCE(Distancefromfiredepartmenttoemergency,whichisspecifiedinkilometers)field.
DISTANCE: available for 50.4% of the listed incidents, and in fact only reported for theincidents from Ontario. However, 45.1% of the values entered are “0”, which may beinterpretedasanactualdistanceshorterthan0.5kmormayrepresentimpropercodingforan undetermined/non‐recorded distance. On the other hand, the remaining values are intheinterval[1,4600],withlargervaluesbeingdoubtfulasactualdistancesinkilometers.
WhiletheNFIDrepresentsasetofrelevantdatafortheanalysisofvariousfactorsassociatedwiththeoccurrenceof fires,areport issuedinSeptember2017bythetheCanadianCentre for JusticeStatistics [4], prepared for the Canadian Association of Fire Chiefs, summarizes the jurisdiction(provinces)thatprovidedincidentdataforvariousNFIDfields–ineffectindicatingotherdatagapsbeyondtheoneswehavereportedabove.
6
FIRESINRELATIONTOOTHEREMERGENCIES
Theproportionoffiresandfirerelatedincidentsinrelationtoothertypesofemergenciesthatfiredepartmentsrespondto,astheresearchteamfound,isperhapsequallyassignificantas,ifnotevenmore critical than, the issues and concerns raised above regarding availability andconsistency/qualityoftheNFIDdataforpurposesofourresearch.TherelevanceandusefulnessofNFID as a national database – one that allows the development of evidence‐based research toenablebetterunderstandingandawarenessof fire incidentsandcreateknowledge for improvingfiredepartmentresponsiveness–becomesdoubtfulwhenconsideringthefactthatfiredepartmentoperations coverwaymore than responding to fire incidents.We initially reviewedTorontoFireServicesoperationsdataasreleasedinJuly2017:
https://www1.toronto.ca/wps/portal/contentonly?vgnextoid=e04015093da69510VgnVCM10000071d60f89RCRD.
InFigures1‐8,wereportonbreakdownsofemergencyincidentsthatwererespondedtobythefiredepartments/servicesoffourcities–Calgary,Ottawa,Toronto,andVaughan–ineachoftheyears2011and2016.Thisissomehowindicativeofevolution,inthecaseofthesefourcities,ofincidentbreakdowns over the most recent five‐year period (2011‐2016). Fire departments’ workloadsclearlydonotcomeexclusively,norevenprincipally,fromfireorfirerelatedincidents,asmaybereadilygleanedfromthesefigures.
7
FIGURE1.CALGARYFIREDEPARTMENT:BREAKDOWNOFEMERGENCYINCIDENTS,2011(DataSource:CalgaryFireDepartmentAnnualReport2011[2])
FIGURE2.CALGARYFIREDEPARTMENT:BREAKDOWNOFEMERGENCYINCIDENTS,2016(DataSource:CalgaryFireDepartmentAnnualReport2016[3])
8
FIGURE3.OTTAWAFIRESERVICES:BREAKDOWNOFEMERGENCYINCIDENTS,2011(DataSource:OttawaFireService2011AnnualReport[7])
FIGURE4.OTTAWAFIRESERVICES:BREAKDOWNOFEMERGENCYINCIDENTS,2016(DataSource:OttawaFireServices2016AnnualReport[8])
9
FIGURE5.TORONTOFIRESERVICES:BREAKDOWNOFEMERGENCYINCIDENTS,2011(DataSource:TorontoFireServices2011AnnualReport[9])
FIGURE6.TORONTOFIRESERVICES:BREAKDOWNOFEMERGENCYINCIDENTS,2016(DataSource:TorontoFireServices2016AnnualReport[10])
10
FIGURE7.VAUGHANFIREANDRESCUESERVICE:BREAKDOWNOFEMERGENCYINCIDENTS,2011(DataSource:VaughanFire&RescueService2011AnnualReport[11])
FIGURE8.VAUGHANFIREANDRESCUESERVICE:BREAKDOWNOFEMERGENCYINCIDENTS,2016(DataSource:VaughanFire&RescueService2016AnnualReport[12])
11
In2011,forinstance,forthesefourcities,thepercentagesoffireandfirerelatedincidentstototalincidents responded to by their fire departments ranged between 3.7% and 13% (see Table 1).Accordingly,thesefiredepartmentsrespondedtomuchlargerproportionsofnon‐fire incidents–whichwould not be reported in theNFID.More significantly, the last two columns of Table 1 –corresponding to the numbers of fire incidents reported in the NFID fields CMA_NAME andCSD_NAME, respectively – indicate that very small proportions of fire department resourceassignmentsactuallyfoundtheirwayintotheNFIDin2011.Thisobservationwouldalsoapplytoallotheryearsinthe2005‐2015timehorizoncurrentlycoveredbytheNFID.
TABLE1.FIREINCIDENTSREPORTEDINNFIDIN2011COMPAREDWITHTOTALINCIDENTSRESPONDEDTOBYFOURCITYFIREDEPARTMENTS
Itshouldbementionedthat,insomeannualreports,thebreakdownofincidentsdoesnotseparateactualstructuralfiresfromotherfirerelatedemergencies(e.g.,vehiclefiresoropenfires).Wheredifferentiation of structural fires is made, a narrower range of 3% to 10% (Fraser InstituteResearchBulletin,2015[6]).Intermsofallocationofresources(firevehiclesandcrewsassigned),responding to fire incidents demandsmore resources than other incident types. For instance, in2016, responses to fire incidents in Toronto corresponded to almost 60% of the unit responses(TorontoFireServices2016AnnualReport[10]).Itisclear,nonetheless,thatresourcesdemandedbytheother,morefrequenttypesofemergencieshaveadirectimpactonafiredepartment’sabilityto respond to fire incidents. For purposes of modelling and simulation of a fire department’soperations,therefore,itisnecessarytoincludeallthecategoriesofincidentsrespondedtobyafiredepartment.
In light of the above‐cited gaps in key operational data (incident location, time of alarm receipt,response time, etc.), as currently reported in the NFID, our research team decided to seek theassistanceoftheVaughanFireandRescueService(VFRS),intheprovinceofOntario,inordertobeabletodevelopasimulationmodelasenvisionedinourproposal.
OuraimwastodevelopafairlygenericmodelthatcouldbereplicatedforfiredepartmentsacrossCanada,foraslongastheappropriatesetofoperationaldataarecollectedandmaintainedbysuchother fire departments.Our case study and the resulting simulationmodelwouldnot have beenpossiblewithouttheassistanceandactiveparticipationofkeyofficialsofVFRS(FireChief,DeputyFire Chief, and a Fire Captain overseeing themaintenance of their incident database). The VFRS
City
Total
Incidents
Fire & Fire
Related Incidents % of Total
in Field
CMA_NAME
in Field
CSD_NAME
Calgary 50,520 1,869 3.7% 1,320 1,190
Ottawa 26,370 3,421 13.0% 1,164 1,126
Toronto 145,334 10,248 7.1% 6,925 3,368
Vaughan 10,166 813 8.0% None 369
No. of Incidents Reported in NFID
12
dataset consisted of operational data covering the years 2009 through 2016. Throughout theconduct of our case study and the associated modelling and simulation (M&S) effort, the VFRSofficials provided clarification and guidance with respect to the interpretation and use of therelevantdatafields.
Theremainingsectionsofthisreportwilldescribetheoverallcasestudy,includingtheinformationthat we processed and the platforms/methods we employed to build the M&S framework. Wedecided to develop the simulation framework taking into account the standard informationavailableinthecaseofVFRSatfiredepartmentlevel(inaccordancewithdirectivesissuedbytheOfficeoftheFireMarshalofOntario).Ourresearchproposalhadanticipatedasufficientlyadequatelevel ofdetail anddata availability in theNFID toallow for ameaningful andproductiveM&Sofincidentoccurrenceandfiredepartmentresponse.
Incident Dataset: Current Case Study
Inthissection,wedescribethedatasetmadeavailablebyVFRS(‘VFRSdataset’or‘VFRSdata’)forpurposesofourcasestudyandM&Seffort.Consistentwith thenon‐disclosureagreement,chartsandtablespresentdataprovidedwiththeconsentofVFRSrepresentatives.
The VFRS data cover eight years of consecutive incident records from January 2009 throughDecember2016.Inordertoaddressfilesizeissues,theVFRSdatasetwasbrokendownandmadeavailabletotheresearchteaminseveralMSExcelworksheets:
a. IncidentMainFeatures,
b. IncidentRespondingUnits,
c. IncidentCivilianCasualties,
d. IncidentFirefighterCasualties,and
e. IncidentOtherTables.
ItispossibletoextractfromtheVFRSdataasetofkeyfeaturesrelatedwithincidentsandresponsecharacteristics.Some fieldscoincidewith thosereported in theNFIDon the incident informationfields, such as INCIDENT ID, ALARM TIME, RESPONSE TIME and INCIDENT LOCATION. SomerelevantfieldsinadditiontothoseinNFIDare:TYPEOFINCIDENTandONSCENETIME.
We initially undertook an assessment of data availability and quality. We sought to eliminatewronglycodedvaluesandoutliers.Oftheabovementionedkeydatafields,theworstcase,forONSCENETIME,provided88%ofutilizabledata(availablerecordsafterthecleaningupprocess).
In Figure 9, we present the percentage distribution of the incident records on an annual basisthroughout2009to2016.Monthlypercentagedistributionthroughout2009to2016ispresentedinFigure10.
13
FIGURE9.VFRSYEARLYINCIDENTS:NUMBERANDPERCENTAGETOEIGHT‐YEARTOTAL(2009‐2016)
FIGURE10.VFRSMONTHLYINCIDENTS:NUMBERANDPERCENTAGETOEIGHT‐YEARTOTAL(2009‐2016)
Byobserving themonthlynumbersof incidentsover2009 to2016 inFigure11, it ispossible todetectaslightupward(growth)trend.
14
FIGURE11.VFRSDATA:TOTALNUMBEROFINCIDENTSPERMONTH(2009TO2016)
InFigure12,weperceiveanoverallupwardtrendinthemonthlynumberofvehiclerescues,withpotentialseasonalcomponents(forinstance,highsorlowsduringcertainmonths).Forfireincidentcalls, the plots in Figure 13 depicts a slight downward (decreasing) trend over the time. ThisdownwardtrendinfireincidentsmaybeindicativeofsuccessfulfirepreventioneffortsbyVFRS.
FIGURE12.VFRSDATA:TOTALVEHICLERESCUESPERMONTH(2009TO2016)
15
FIGURE13.VFRSDATA:TOTALFIREINCIDENTSPERMONTH(2009TO2016)
Fromouranalysisof time intervals (inminutes)betweenconsecutivevaluesofALARMTIME (or“inter‐arrivaltimes”),weobtaintheinter‐arrivaltimedistributionforemergencycallsinFigure14,the frequency histogram for the overall incident list represents the expected inter‐arrivaldistribution,whichsuggestsanegativeexponentialfunctionandwhichisconsistentwithaPoissondistributionof“arrivals”ofemergencycalls.(ThePoissondistributioncommonlycharacterizesthearrivalofcustomersinaservicequeuingsystem.)
FIGURE14.VFRSDATA:DISTRIBUTIONOFTIMEBETWEENSUCCESSIVECALLS,INMINUTES
16
Figure15showsthedistributionofRESPONSETIMEintheVFRSdataset.Basedonusualexperienceinservicesystems,a‘moundshaped’–ornormal–distributionisgenerallyexpected,whichimpliesthattheprocessismatureandwellimplemented(i.e.,theprocessandorganisationhavepassedthelearningcurveeffect).
FIGURE15.VFRSDATA:RESPONSETIMEDISTRIBUTION(ALLINCIDENTTYPES)
However,weapplyinoursimulationstudiestheactualempiricaldistributionsofRESPONSETIMEaccording to various incident types.We analysed theRESPONSETIME observed for each type ofincident,obtainingdifferentvaluesofmeansandvariabilitystatistics.
Asimilaranalysis isrequiredforONSCENETIME.Figure16showstheoveralldistributionofONSCENETIME,althoughwefoundparticulardistributions(differentshapesandparameters)acrossvariousincidenttypes.
17
FIGURE16.VFRSDATA:ONSCENETIMEDISTRIBUTION(ALLINCIDENTTYPES)
Besidesthedistributionsofinter‐arrivaltimes,responsetimes,andonscenetimes,anotherhighlyrelevant component forM&S of emergency calls is their spatial distribution. The Longitude andLatitude coordinates recorded for each reported incident allow us to capture spatial patternsbehindvariousincidenttypes.WecreatedapartitionoftheentiregeographicalregioncoveredbyVFRS,usingalatticegranularityof500meters×500meters(0.25km2).‘Heatmaps’,asdepictedinFigure17,showthespatialanalysisfor:
a) Allincidenttypesin2009;
b) Allincidenttypesin2009‐2016;and
c) Vehiclecollisionincidentsin2009‐2016.
Thevaluesappearingineachcellarethenumbersofaccumulatedincidentswhichhaveoccurredinthat area over the period specified. Each incident type produces a different spatial distributionpattern(i.e., ‘hotspots’locatedinspecificareasdependingontheincidenttype),whichisrelevantforourM&Sframework.
18
FIGURE17.VFRSDATA:SPATIALDISTRIBUTIONANALYSISFORSPECIFICINCIDENTTYPES(2009‐2016)
Inadditiontothespatialdistributionpatternsresultingfromthelongitudeandlatitudeinformationforeachincident,itispossibletoestimatetheexpectedtraveldistancefromtherespondingstationtotheincident.
We are also able to obtain, for each reported incident, an Alarm Processing Plus TurnOut Time(APPTOT), which is calculated as Roll‐out Time stamp (when the vehicle rolls out of its homestation)minusAlarmReceiptTimestamp(whenthealarmisreceivedattheVFRScommunicationscentre). Thedistribution ofAPPTOT is presented in Figure 18.Wemust point out that theVFRSdatasetonly includesasdata fieldstheAlarmReceiptTimestamp(alm_time intheIncidentMain
19
Features file, or notif_time in the Incident Responding Units file) and the Roll‐out Time stamp(roll_time in the Incident Responding Units file). We use the Roll‐out Time stamp of the firstrespondingunitatthesceneincalculatingtheAPPTOT.Sincetheoriginatingstationisnotclearlyidentifiedbywayofaseparatedatafield,weassumethatthefirstrespondingunitoriginatesfromthestationresponsible for thegivenEmergencyPoint.AlarmProcessingTimeandTurnoutTimearenotseparatelyrecorded.
FIGURE18.VFRSDATA:APPTOTDISTRIBUTION(ALLINCIDENTTYPES)
OurM&Sframeworkinvolvedtwodifferentsimulationmodelsrunningonseparateplatforms:
1. AnIncidentGenerationEngine,developedandimplementedusingCPNTools4.0[5],whichsimulatesthe‘arrival’ofincidents.Eachincidentoccurrenceischaracterizedby:
a. typeofincident,
b. location,
c. ‘arrival’ time (based on the time stamp of Alarm Receipt at the VFRScommunicationscentre),
d. APPTOT,and
e. onscenetime.
2. AResponseSimulationModel,developedandimplementedusingAnyLogic8.0.5[1].
20
Thesetwoseparatesimulationmodelsarefullydescribedinthetwoimmediatelysucceedingtwosectionsofthisreport.
Incident Generation Engine
In this section, we describe the first of two simulation models. This Incident Generation Engineproduces a list of incidents (including the main features: type, location, ‘arrival’ time), which isgeneratedbasedontheempiricaldistributionsobtainedforthekeyincidentfeatures.ThisincidentlistprovidestheinputsfortheResponseSimulationModel.
AconceptualoverviewoftheintegrationoftheinformationproducedfromtheanalysisoftheVFRSdatasetispresentedinFigure19.
FIGURE19.INTEGRATIONOFEMPIRICALDATAFORSIMULATIONOFTHEINCIDENTLIST
Thesimulationof incidentoccurrenceshasbeendevelopedand implementedasadiscreteeventsimulation model with CPNTools 4.0 [5] as the platform. The model takes simultaneouslyinformationfromtheempiricaldistributions,consideringcentraltendencyandvariabilitystatistics,ofvariousdatafieldswithinthedatasettogenerate:
1. type of incident, considering the probabilities of occurrence of specific types of incidentsover time (at the level of frequencies on an hourly basis during the day), and at fourdifferenttimewindows(TWs)duringtheday(i.e.0:01‐5:59h,6:00‐11:59h,12:00‐17:59hand18:00‐23:29h.),
2. incidentlocation,basedonthe500m×500mcellsinthelatticeaspreviouslydescribed,
3. incident ‘arrival’ time,basedon thedistributionof inter‐arrival timesbetweensuccessivealarms,
21
4. APPTOTforthefirstrespondingcrewdependinguponthetypeofincident,and
5. onscenetimeforthefirstrespondingcrew,dependinguponthetypeofincident.
The generation of incidents for one day has been implemented using colored Petri nets withCPNTools 4.0 [5] as platform. An overview of the main components of the model (places,transitions,andarcs)ispresentedinFigure20.
FIGURE20.INCIDENTGENERATIONENGINE:MAINCOMPONENTS
Theempiricaldistributionsobtained fromtheVFRSdatasetare themain inputsprovided for thesimulationthatproducestheincidentlistforanentireday.Descriptionsofcomponents(networknodes)aresummarizedinTable2.
Theoutputlistincludesthefollowingfeatures/attributesforeachincident:
ID:incidentID
TP:incidenttype
ST:spatiallocation(cellinthelattice)
t1:inter‐arrivaltime
PT:APPTOT
OT:onscenetime
22
TABLE2.DESCRIPTIONSOFINCIDENTGENERATIONENGINECOMPONENTS(PLACESANDTRANSITIONS)
Name Nodetype Description
Incidentmain TransitionTakesonevaluefromeachoftheplaces(nodes)connectedtosimulatethemainfeaturesforanincident(interarrivaltime,type,location).
APPTOT TransitionTakesonevaluefromthedistributiondependingontheincidenttype.
Onscenetime TransitionTakesonevaluefromthedistributiondependingoftheincidenttype.
Day PlaceContainstheinformationoftheday(Monday,Tuesday,etc.),Month(Jan,Feb,etc.)forthegenerationoftheincidentlist.
Inter‐arrivaltime PlaceContainstheinter‐arrivaltime,obtainedfromthedistributionoftimebetweensuccessivecalls.
Typedailydistribution PlaceContainstheincidentoccurrencedistributionperdayforeachincidenttype.
Typemonthlydistribution PlaceContainstheincidentoccurrencedistributionpermonthforeachincidenttype.
Typehourlydistribution PlaceContainstheincidentoccurrencedistributionperhourforeachincidenttype.
Typeoveralldistribution PlaceContainstheincidentoccurrencedistributionforeachincidenttype.
Typelocationdistribution PlaceContainstheincidentoccurrencedistributionperlocation(cellsof0.5km×0.5km)foreachincidenttype.
Timewindows PlaceContainstheincidentoccurrencedistributionwithinthespecifiedTWsforeachincidenttype.
Incidentcall PlaceContainstheincidentsgeneratedalongwiththemainfeatures(interarrivaltime,type,location)
APPTOTdistribution PlaceContainstheestimatedAPPTOT,obtainedfromthedistributionfortheoverallincidenttypes.
Incidentresponse PlaceContainstheincidentsgeneratedalongwiththemainfeaturesplusthepreparationtime.
Onscenetimedistribution PlaceContainstheincidentonscenetimedistributionforeachincidenttype.
Incidentcleared PlaceFinalnodewithallthefeaturessimulated:interarrivaltime,type,location,APPTOT,onscenetime.
23
Response Simulation Model
TheincidentlistcreatedusingtheIncidentGenerationEngineservesasinputintothesecondmodel–theResponseSimulationModel.InordertorecreatetheenvironmentinwhichVFRSrespondstoemergencies of various types occurring throughout the day, we developed and implemented anagent‐basedsimulationmodelusingtheAnyLogic8.0.5simulationplatform[1].
While the IncidentGenerationEngine isbasedondiscreteeventmodelling,weusedAgent‐BasedModelling (ABM) to develop our Response Simulation Model. ABM is best described as adecentralized, individual‐centric approach tomodelling. In this approach, individual participantshavetheirownbehaviorandarereferredtoasagents.
WithintheAnyLogicsimulationplatform,weareabletotransformreportedlongitudeandlatitudeinformationonincidents(aswellasfirestations)intousableGeographicInformationSystem(GIS)coordinateswhichallowsimulationofvehiculartraveloncitystreets.
MODELDESCRIPTION
Theagentsthatcomeintoplayinthemodelare:
Dispatcher–thepersonwhoreceivesthe911callfromtheEmergencyPointanddecidestoalerttheappropriateStationtoattendtotheincident;
EmergencyPoint–anentitythatdevelopsandchangesitsstatusbasedontheactionsoftheotheragents;
Station– theagent thatreceives theexecutionorder fromtheDispatcherandchanges itsstatusaccordingtoavailabilityofresources;and
Vehicle – the entity that receives thedispatchingorder from the available Station. It alsousestheGISMapinordertotaketheappropriateroute.(BasedonVFRSpractice,acrewoffourfirefightersmanseachrespondingvehicle.)
FIGURE21.AGENTSUSEDINTHEMODEL
24
As faras thesimulationmodelanimationview is concerned,a scaledrepresentationof thecity’sFireDistricts and stations involved helps the user to directly visualizewhat is happening in theenvironmentintermsofentitiesandtheflowofresources(seeFigure22).
FIGURE22.SUBDIVISIONOFTHECITYOFVAUGHANINTOFIREDISTRICTSANDCURRENTFIRESTATIONLOCATIONS
TheGISMapshapeenablesonetodisplayandmanageGIS(GeographicInformationSystem)mapsinthemodel.TheGISRegionelementtomarkupsomeclosedareaonthemapwasused,anditispossibletoaddanewregionormodifytheregionshape.Eachpointoftheregionhaslatitudeandlongitudecoordinates,definedindegrees.Itisalsopossibletouseashapefile(.shp)asaninputtobuildtheregioninsidetheCityDistrict.Thedefinitionofregionsisveryimportantbecausethelogicinside themodel is structuredondifferentdecision levels basedon theseboundaries. The agentthat is charged to take these decisions is theDispatcher that receives the latitude and longitudeinformationfromtheincident.Usingthesevaluesitispossibletodeterminewhichregion(district)andstationtakesprincipalresponsibilityfortheincident.
IntheinitialResponseSimulationModeldevelopmentandtesting,ourmodelwasrestrictedtooneresponding vehicle per incident. According to currentVFRSprotocol, different types of incidentscall for 1, 2, or 4 responding vehicles. In applying current protocol, we obtained the followingdistributionofnumberofrespondingunitsrequired:
25
Thus,only5.3%ofincidentswouldbeexpectedtorequirefourrespondingvehicles.Wealsolookedinto the actual numbers of responding units, as reported in the Incident Responding Units file(January2009–December2016)andderivedthefollowingdistribution:
Basedonactualnumbersofunitsrespondingtoemergencyincidentsin2009‐2016,89.7%involvedoneortworespondingvehicleswhileonly10.3%involvedmorethantwovehicles.
Thus,whether based on current protocol or on experience,we see a relatively small percentage(5.3%and10.3%,respectively)of incidentsrequiringmorethantworespondingunits.Our teamdecided that a Response Simulation Model involving either one or two responding units wouldenable a relatively simple working model, but one which still is roughly indicative of resourcerequirements.Accordingly,allincidenttypecodesrequiringeithertwoorfourrespondingvehiclesperVFRSprotocolare–inourResponseSimulationModel,ascurrentlydeveloped–assignedtworespondingunits.
Currently,themodelusesthefollowinglogictosendresourceswheretheyareneeded:
1. Considerthestationsthatareinthesameregionastheincident.
2. ChoosetheNearestStationinRegionbyRoute.
3. Checktheavailabilityofresources.
4. ChecktheresourcesneededbytheEmergencyPoint.
5. If the available Station resources are exactly the number of needed resources ofEmergencypoint,sendallvehiclesrequired.
6. If not, send the vehicles available in the Station and update the resources neededdecreasingthevalue.
Number of Trucks per Protocol %‐age
1 60.5%
2 34.1%
4 5.3%
100.0%
Vehicles responding # incidents % occurrence
1 57,494 69.4%
2 16,819 20.3%
3 2,864 3.5%
4 1,712 2.1%
More than 4 3,909 4.7%
82,798 100.0%
26
7. If resources needed to Emergency Point is greater than zero, check other Stations inRegionand:
7.1 If the available Station resources are exactly the number of needed resources ofEmergencypoint,sendallvehiclesrequired.
7.2 If not, send the vehicles available in the Station and update the resources neededdecreasingthevalue.
8. IfresourcesneededbyEmergencyPointisgreaterthanzero,choosetheNearestStationbyRouteamongallothersintheentirecitywithavailableresourcesand:
8.1 IftheavailableStationresourcesareexactlythenumberofneededresourcesofEmergencyPoint,sendallvehiclesrequired.
8.2 Ifnot,sendthevehiclesavailableintheStationandupdatetheresourcesneededdecreasingthevalue.
Itisalsopossibletochangethisdecision‐makingschemeinconnectionandconsistentwithcurrentoperatingprocedures/protocols as used in a given city. In order to better understand thewholelogicthatisinthemodel,apseudo‐algorithmhasbeenwrittenbelowthatcanrepresentthemainactionsandthedecisionprocessstepbystep.
inputfile:EmergencyPoints(IdentificationCode,Latitude,Longitude,Timestamp,APPTOT,Onscenetime,EmergencyCode,numberofvehiclesneeded)readFileperrowcalculateinwhichRegiontheEmergencyPointisinputfile:Stations(StationID,Latitude,Longitude,Regionofbelonging,NumberofVehicles)readFileperrowSetStationinGISMap;for(eachEmergencyPoints.file.row){if(themodeltimeisequaltoincidenttime)EmergencyPoint.statuschangestoActive;EmergencyPointsendthemessage:alarm;Dispatcherreceivesmessage;Dispatcherreceivesinformation:EmergencyPointparameters;SelectStationsinRegion;ChoosetheNearestStationinRegionbyStreetDistance;if(ResourcesAvailable&&resourcesneeded>0)Dispatchersendthemessage:operation;Stationreceivesmessage;Station.StatuschangestoOperation;Stationsendthemessage:go;Vehiclereceivesthemessage;Vehicle.statuschangestoPreparation;Station_selected.numberVehicles‐‐;resourcesneeded‐‐;Vehiclegoesto(EmergencyPoint.Latitude,EmergencyPoint.Longitude)else(ArethereotherStationsinRegion?)if(ResourcesAvailable&&resourcesneeded>0)Dispatchersendthemessage:operation;Stationreceivesmessage;Station.StatuschangestoOperation;
27
Stationsendthemessage:go;Vehiclereceivesthemessage;Vehicle.statuschangestoPreparation;Station_selected.numberVehicles‐‐;resourcesneeded‐‐;Vehiclegoesto(EmergencyPoint.Latitude,EmergencyPoint.Longitude)elseif(resourcesneeded>0)(identifyavailableStations.outsideRegion)ChoosetheNearestStationbyStreetDistance;Dispatchersendthemessage:operation;Stationreceivesmessage;Station.StatuschangestoOperation;Stationsendthemessage:go;Vehiclereceivesthemessage;Vehicle.statuschangestoPreparation;Station_selected.numberVehicles‐‐;resourcesneeded‐‐;Vehiclegoesto(EmergencyPoint.Latitude,EmergencyPoint.Longitude)activatefunctionCheck.Station.Availability;if(numberVehicles==0)Available=false;Station.StatuschangestoUnavailable;Vehicle.statuschangestoOnRoute;Agent.Vehiclearrival‐>sendthemessage:OnSite;EmergencyPointreceivesthemessage;EmergencyPoint.statuschangestoRescue;Time=OnSceneTime;EmergencyPoint.statuschangestoCleared;Vehicle.statuschangestoReturning;Vehiclegoesto(Station.Latitude,Station.Longitude)Agent.Vehiclearrival‐>Station_selected.numberVehicles++;activatefunctionCheck.Station.Availability;
}
With an agent‐basedmodel,wepopulate the environmentwith the various entities that interactwith each other. As mentioned previously, the agents used are the incidents, the vehicles, thedispatcher,andthestations.Thestationsareplacedinthemapbyuploadinga file;hence, inthismodel,anyactualnumberofstationsmaybespecified.Eachstation’sparameterswill includethefollowinginformation:
StationID;
Latitude;
Longitude;
Regiontowhichthestationbelongs;and
Numberofvehicles.
Regarding the number of vehicles per station, it is possible to set it at the beginning of thesimulationthroughappropriateslidersorbyenteringthenumberdirectlyinabox(seeFigure23).
28
FIGURE23.MODELHOMEPAGEANDSETTINGNUMBERSOFAVAILABLEVEHICLES
The operational capability of the stations depends on the availability of resources. The greensymbolontopofastationmeansthattheagent(station)isabletorespondtoanincident.Whenallthevehiclesbelonging to thatstationhavebeendispatchedor theyarealready inpreparation toattendtoanincident,thestationisunavailableintermofresources.Forthepurposeofshowingthestationunavailability, theblacksymbolon topof thestationwasdesigned,and itmeans that theagent isnotabletorespondtoany incident.Figure24showsthesymboldescribedpreviously inconnectionwithstationstatus.During thisstateofunavailabilityof thestation, it is important todefinethelogicofimplementationofthemodelandcollecttheoutputdataappropriately.
FIGURE24.STATIONSTATUSANDRELATEDSYMBOLS
For each station a certain number of vehicles is assigned. Each vehicle is an agentwith its owncharacteristicsandbehaviors.Tobetterunderstandhowthesetypesofagentsbehaveinthemodel,thestatusofeachvehiclehasbeensummarizedinthefollowinglist:
29
Waitingisthestatusinwhichthevehiclewaitsatitsstation.Thisstatemeansthatthevehicleisavailableforassignmenttoanincident.
PreparationstatusstartsassoonasthealarmisreceivedbytheVFRScommunicationscentreand lastsuntil thevehiclerollsoutof thestation.SinceAlarmProcessingTimeandTurnoutTimearenotseparatelyrecorded,themodeltreatsavehicleasbeinginthe‘Preparation’statusthroughouttheperiodoftimereferredtoasAPPTOT,asprovidedbytheIncidentGenerationEngine.(AstationandavehicleareassignedtoanincidentbaseduponthethelongitudeandlatitudeinformationoftheEmergencyPoint.)
ThestatusOnRoutereflectsthevehicletravelingtowardstheEmergencyPoint. TheOnScenestatusreflectsthevehiclebeingattheEmergencyPointandattendingto
theincident. Returningisthestatusaftertheemergencyhasbeencleared,withthevehicletraveling
backtoitshomeStation.
Thevehiclesreachthe incident locationusingtheexistingroadsandroutesbasedonrealspatialdata.Furthermore, thankstoGISMapfeatures, thesimulatorchooses the fastestway toarriveattheincident.Asalreadymentionedabove,anotherextremelyrelevantentitywithinthemodelistheEmergencyPoint,withthefollowingcharacteristicsasspecifiedbytheIncidentGenerationEngine:
Incidentidentificationcode; Latitude; Longitude; AlarmReceipt Time stamp (when the alarm is received at the VFRS communications
centre); APPTOT; Onscenetime;and Incidenttypecode.
The Emergency Point information is entered into the model as an input file, where each rowrepresents an incident. The incident type codeautomaticallydetermines thenumberof requiredvehiclestobeassignedtorespondtoagivenincident.SincethesimulatorallowstheuploadoftheinputfilecomingfromtheIncidentGenerationEngine,therearenorestrictionsonthenumberofemergenciesthatcanbesimulatedinarun.
TheEmergencyPointagentiscloselyrelatedtotheotheragents’statusdiagram,becauseitsendsandreceivesinformationto/fromalltheotheragentsinthemodel.Thedifferentstatesrelatedtoanemergencypointarelistedbelow:
Inactive:Thestateinwhichtheincidenthasnothappenedyet; Active: The state inwhich the Emergency Point sent an alarm and it iswaiting for a
response; Rescue:StateinwhicharesponderhasarrivedattheEmergencyPoint; Cleared:ThestateinwhichtheEmergencyPointhasbeencleared.
Byusingstatechartsonecanvisuallycaptureawidevarietyofdiscretebehaviors.,Forthisreason,statecharts were used to better represent the various agent states in the model (as described
30
above).Figure25 includes theagentsstatecharts thatweredeveloped in themodel toreflect therealenvironmentprocess:
FIGURE25.AGENTS’STATECHARTS
Thesimulationmodel,asdeveloped,providesasetofindicatorsofoperationalperformance.Suchindicatorsaresimplyvariousperformancemeasuresrelatedtothedifferentrescuesoperations.Tofulfilltheirpurpose,suchindicatorsaresimpletounderstand.Oncethesimulationiscompleted,abuttonbecomesavailableinthewindow,toallowtheusertocollectthesimulationresultsavailableintermsofoutputfiles.
31
Theoutputfilescontainthefollowingperformanceindicators:
Timeperiodbetweenalarmreceiptandtheendoftherescueprocedures,
Traveltimeofeachvehicleusedandthetotaltimeamountofallvehiclesperday,
Traveldistancesofeachvehiclesintermsofmetersandthetotalamountofallvehicles
perday,
Accumulatednumberoftheincidentsoverthetime,
NumberofvehiclesremainingperStation,
Vehicles sent by other Stations for each Emergency point,when the Station thatwas
supposedtorescuehasn’tallvehiclesneededavailable.
Thesimulationmodelhomepageallowstheusertochoosebetweentwooptions:
1. immediately launch the simulation runwithout worrying aboutmodel customizationandagreeingwiththedefaultconditionsandsettings;or
2. modify pre‐implemented default settings and customize themodel input parameters,enablingthemodeltotestdifferentconfigurations.
Thehomepagecontainsdifferentnumberofboxwheretheusercansetupthemodelparametersrelatedtoagentinthesimulatedenvironment,asshownpreviously(Figure23).Eachrunincludesthesimulationofadayofactivityof the firedepartment,duringwhichvariousemergenciesmayoccuraccordingtothestatisticaloutputsdefinedintheprevioussection.Figure26showsaviewofthe performancemeasures listed above that the simulationmodel is able to provide during thesimulationandalsocollectinadataset.
FIGURE26.PERFORMANCEINDICATORSVISUALIZEDDURINGSIMULATION
32
SIMULATIONRESULTS(FUNCTIONALITYVALIDATIONOFTHEMODEL)
EachreplicationofoursimulationexperimentsimulatesonedayofVFRSresponsestoemergencyincidents.Wehavethuscompletedandcollectedstatisticsfrom180replications(correspondingtoasix‐monthperiodfromJanuarytoJune)oftheexperiment. Ineffect,180daysof incidentsweregeneratedusingtheIncidentGenerationEngine,andtheresultingincidentlistwasusedasinputfortheResponseSimulationModel.
Wepresentthedistributionsofresponsetimesonamonthlybasisaftereliminatingoutliers(nearly5% percent of the simulated values). All the presented values were obtained by simulating theresponse times with empirical distributions derived from the Incident Generation Engine incombination with the Response Simulation Model. Figures 27‐33 present the response timedistributionsobtainedforeachmonthfromJanuarytoJune.
FIGURE27.DISTRIBUTIONOFSIMULATEDRESPONSETIMES,JANUARY
33
FIGURE28.DISTRIBUTIONOFSIMULATEDRESPONSETIMES,FEBRUARY
FIGURE29.DISTRIBUTIONOFSIMULATEDRESPONSETIMES,MARCH
34
FIGURE30.DISTRIBUTIONOFSIMULATEDRESPONSETIMES,APRIL
FIGURE31.DISTRIBUTIONOFSIMULATEDRESPONSETIMES,MAY
35
FIGURE32.DISTRIBUTIONOFSIMULATEDRESPONSETIMES,JUNE
FIGURE33.DISTRIBUTIONOFSIMULATEDRESPONSETIMES,JANUARY–JUNE
ByinspectionofFigures27–33,thedistributionsobtainedshowjustslightlydifferentshapesfromonemonth to the next. Our simulation experiment yielded an average travel speed of 41 km/hrbasedon the travel time(ArrivalTimestampminusRolloutTimestamp)and thestreetdistancefromtherespondingstationtotheincidentlocation.
36
As part of model validation, we summarize in Table 3 the distribution of simulated values ofresponsetimes,andcomparethestatisticswiththoseofactualvaluesoverallin2009‐2016:
TABLE3.STATISTICALCOMPARISONOFSIMULATEDVS.ACTUALRESPONSETIMEDISTRIBUTIONS
Onaverage,oursimulationyieldslongerresponsetimes(thedifferencebetweenoverallmeansisslightlyover37seconds),andexhibitinggreatervariability.Thismayindicatetheneedforamoresophisticatedmodelforrespondingvehicles’speedsforincidentslocatedinthemore‘rural’areas(withlongertraveldistancesandpossiblyhigheraveragespeedascomparedwiththemoreurbanareas).
Besidestheanalysisoftheresponsetime,themodelallowsthegenerationandcollectionofotherrelevantoperationalindicators(e.g.traveldistance/time,workloaddistribution,etc.).InFigure34,forinstance,wepresenttheaveragetotaltraveldistanceforfirstrespondingunits.ThesimulateddistributionofincidentsamongthedifferentregionswithinthecityofVaughanisshowninFigure35. In Figure 36, we see the breakdown of responding vehicle/crew overall time into APPTOT,Travel,andOnscenetimes.
MonthMean
(minutes)StdDev(minutes)
Max(minutes)
Min(minutes)
n
January 6.31 3.17 16.8 1.2 873February 6.42 3.26 16.8 1.5 796March 6.29 3.19 16.9 1.6 891April 6.39 3.2 16.9 1.3 884May 6.44 3.17 16.9 1.4 885June 6.29 3.03 16.9 1.3 904
Overall(180replications) 6.36 3.17 17 1.2 5,233Overall(actualvalues) 5.74 2.32 17 1 80,310
37
FIGURE34.AVERAGETRAVELDISTANCEPERINCIDENT(INMETERS)FORFIRSTRESPONDERVEHICLES(180REPLICATIONS)
FIGURE35.SIMULATEDNUMBEROFINCIDENTS,BYREGION(180REPLICATIONS)
38
FIGURE36.RESPONDINGVEHICLE/CREWOVERALLTIMEBREAKDOWN,BYREGION(180REPLICATIONS)
Conclusions and Further Research
The current Incident Generation Engine involves stochastic modelling and simulation of theoccurrenceofincidents(asspecifiedintermsofincidenttype,‘arrival’time,location,APPTOT,andon scene time). In spite of the inherent variability/uncertainty in observed values, our incidentgenerationmodelappearstoprovidesimulatedresponsetimesthatmoreorlessfollowtheactualpatterns arising in the historical data.However, themodel still needs to be calibrated to furthervalidateresultingdistributionsofincidentattributesinrelationtoactualdistributions.
Thesimulationproducedbymeansof thetwomodels(IncidentGenerationEngineandResponseSimulation Model) as developed is capable of reproducing the performance of the system.Deviationsobservedfortheexperiment(180replicationseachhavingadurationofoneday)comemainly from thenecessity of calibrating themodel (for instance, fine tuningof someparameterssuchasaveragevehiclespeed).
The Response SimulationModel is intended to provide a flexible tool for planners and decisionmakerstoevaluatevariousoperatingscenariosoralternativeprocedures/protocols–e.g.,differentnumbers, types or allocation of resources, expected demand fluctuations, alternative assignmentrules,amongothers.
39
EXTENSIONANDIMPROVEMENTOFTHEDEVELOPEDMODELS
We are currently concluding an extended experiment consisting of 365 replications (or theequivalent of one year of emergency incidents and response operations) and will subsequentlyreportonresults.
Basedontheinitialresultsobtained,furtherworkmayincludeextendingthesimulationmodelstoenable:
DefinitionintheResponseSimulationModelofdifferenttypesofrespondingvehicles,andestablishing protocols for assignment of different types/numbers of responding unitsdependinguponthetypeofincident.
Developing more extensive experiments to evaluate different scenarios – e.g., impact ofresource demand fluctuations on response times/capacity, comparison betweenassignmentofrespondingunitsaccordingtoresponsibledistrict/regionvs.geographicallycloseststation(s).
Acknowledgment
ThisworkwassupportedbyagrantfromtheCanadianAssociationofFireChiefs(CAFC).
The research teamwishes to acknowledgepartial financial support fromProfessor JianhongWu,UniversityDistinguishedResearchProfessorandCanadaResearchChairinIndustrialandAppliedMathematics,ofYorkUniversity.
We are deeply indebted to Chief Larry Bentley, Deputy Chief Deryn Rizzi, Deputy Chief AndrewZvanitajs,andCaptainKevinPlestedoftheVaughanFireandRescueService,notonly formakingavailable their incidentdatasetover theeight‐yearperiod fromJanuary2009toDecember2016,butmostespeciallyforprovidingtheirassistance,guidance,andinsightsinourstatisticalanalysisof the VFRS dataset and the necessary verification/validation/accreditation of our simulationmodels.
References
1. AnyLogicSimulationSoftware,2017.[Downloadableathttp://www.anylogic.com].
2. CalgaryFireDepartmentAnnualReport2011.
3. CalgaryFireDepartmentAnnualReport2016.
40
4. CanadianCentrefor JusticeStatistics,2017.FireStatistics inCanada,SelectedObservationsfrom the National Fire Information Database 2005 to 2014. Report prepared for theCanadianAssociationofFireChiefs,September2017.
5. CPNTools,2017.[Downloadableathttp://cpntools.org/].
6. FraserInstituteResearchBulletin.MunicipalFireServicesinCanada:APreliminaryAnalysis,May2015.
[Available online at https://www.fraserinstitute.org/sites/default/files/municipal‐fire‐services‐in‐canada.pdf(accessedonDec22,2017).]
7. OttawaFireServices2011AnnualReport.
8. OttawaFireServices2016AnnualReport.
9. TorontoFireServices2011AnnualReport.
10. TorontoFireServices2016AnnualReport.
11. VaughanFire&RescueService2011AnnualReport.
12. VaughanFire&RescueService2016AnnualReport.
Author Biographical Information
The research project team is composed of researchers who are affiliated with the School ofAdministrative Studies and the Advanced Disaster, Emergency and Rapid Response Simulation(ADERSIM) programat YorkUniversity. TheDisaster andEmergencyManagement graduate andundergraduateprogramsofYorkUniversityarehousedintheSchoolofAdministrativeStudies.
Dr.AdrianoO.SolisisanAssociateProfessorofLogisticsManagementandDecisionSciencesandcurrently theDirectorof theSchoolofAdministrativeStudiesatYorkUniversity.He teachesandconducts research in the areas of operations/logistics/supply chain management and decisionsciences.HeisamemberofthegraduatefacultyinDisasterandEmergencyManagement,andisanactive collaborator in the ADERSIM program. His research interests are in inventory systemsmodeling, supply chain management, intermittent/lumpy demand forecasting, applied modelingand simulation, and IT in operations and supply chainmanagement. He has served as ProgramChair of the Summer Computer Simulation Conference of the Society forModeling & SimulationInternational (SCSC 2013, Toronto, Canada) and as General or Program Co‐Chair of theInternationalConferenceonModelingandAppliedSimulation(MAS2014,Bordeaux,France;MAS2015,Bergeggi, Italy;MAS2016, Larnaca, Cyprus;MAS2017,Barcelona, Spain).Hewill serve asGeneralCo‐ChairofMAS2018inBudapest,Hungary(September2018).ProfessorSolisholdsBScandMScdegreesinmathematics,anMBA,andaPhDinManagementScience.BeforejoiningYorkUniversity, he was Associate Professor of Operations and Supply Chain Management at theUniversityofTexasatElPaso.HehasbeenaVisitingProfessorattheUniversityofCalabria(Italy),wherehehascontinuedtosupervisemanagementengineeringgraduatestudentsintheirresearch
41
work.His teaching isenhancedby industryexperience, includinghavingbeenVicePresidentandDivision Manager for Professional Products and Systems in the Philippine operations of PhilipsElectronics,theNetherlands‐basedmultinationalcompany.
Dr. Ali Asgary is an Associate Professor of Disaster and Emergency Management in YorkUniversity’sSchoolofAdministrativeStudies.HeisaPrincipalInvestigatorandProgramLeadforYork University’s ADERSIM program. He is an expert in disaster, emergency, and businesscontinuitymanagement.Hisextensiveresearchandeffective teachingareenhancedbyhisactivecontributionstotheprofessionandbytranslatingthemintorealworldpracticesatdifferentlevels.HisresearchhasbeenfundedbySSHRC,NSERC,GEOIDE,PreCarN,AIF,YUFA,andothers.Heistheauthor or co‐author of numerous scholarly and practitioner articles in disaster and emergencymanagement, and his work has been extensively cited and referenced by other researchers.ProfessorAsgaryhas receivedvariousawards forhis research, teachingandother contributions,including the International Association of Emergency Management Award and the outstandingpaperof theyearawardby the JournalofDisasterPreventionandManagement.HeobtainedhisPhD at University ofNewcastle UponTyne in England.Hewas one of the facultymemberswhoestablishedthedisasterandemergencymanagementdisciplineinCanadianuniversities,includingYorkUniversityandBrandonUniversity.
Dr.JenaroNosedalisaPost‐DoctoralVisitoratYorkUniversity’sSchoolofAdministrativeStudiesanditsDisasterandEmergencyManagementprogram,andanADERSIMcollaborator.Heiscoursedirector in Operations Management and Quantitative Methods in Business at the School ofAdministrative Studies. He holds bachelor’s andmaster’s degrees in Industrial Engineering, andreceived his PhD in Telecommunications and Systems Engineering from the AutonomousUniversityofBarcelonain2016.Dr.Nosedal’sgeneralareaofresearchisinoperationsandlogisticsmanagement,specializinginsystemsmodelingandsimulation.
Ms.BeatriceZaccaro holds a Laurea in IngegneriaMeccanica (bachelor’s degree inMechanicalEngineering)andiscurrentlyastudentintheLaureaMagistraleinIngegneriaMeccanica(master’sdegreeinMechanicalEngineering)programattheUniversityofCalabriainItaly.InFall2017,shewas an International Visiting Research Trainee (IVRT) in the graduate program in Disaster andEmergency Management at York University. Her research interests include modeling andsimulationwithapplicationsinlogisticsandinsafetyandsecurityinindustrialplants.
List of Tables and Figures
TABLES
TABLE1.FIREINCIDENTSREPORTEDINNFIDIN2011COMPAREDWITHTOTALINCIDENTSRESPONDEDTOBYFOURCITYFIREDEPARTMENTS 11
TABLE2.DESCRIPTIONSOFINCIDENTGENERATIONENGINECOMPONENTS(PLACESANDTRANSITIONS)22
TABLE3.STATISTICALCOMPARISONOFSIMULATEDVS.ACTUALRESPONSETIMEDISTRIBUTIONS 36
42
FIGURES
FIGURE1.CALGARYFIREDEPARTMENT:BREAKDOWNOFEMERGENCYINCIDENTS,2011 7
FIGURE2.CALGARYFIREDEPARTMENT:BREAKDOWNOFEMERGENCYINCIDENTS,20167
FIGURE3.OTTAWAFIRESERVICES:BREAKDOWNOFEMERGENCYINCIDENTS,2011 8
FIGURE4.OTTAWAFIRESERVICES:BREAKDOWNOFEMERGENCYINCIDENTS,2016 8
FIGURE5.TORONTOFIRESERVICES:BREAKDOWNOFEMERGENCYINCIDENTS,20119
FIGURE6.TORONTOFIRESERVICES:BREAKDOWNOFEMERGENCYINCIDENTS,2016 9
FIGURE7.VAUGHANFIREANDRESCUESERVICE:BREAKDOWNOFEMERGENCYINCIDENTS,2011 10
FIGURE8.VAUGHANFIREANDRESCUESERVICE:BREAKDOWNOFEMERGENCYINCIDENTS,2016 10
FIGURE9.VFRSYEARLYINCIDENTS:NUMBERANDPERCENTAGETOEIGHT‐YEARTOTAL(2009‐2016) 13
FIGURE10.VFRSMONTHLYINCIDENTS:NUMBERANDPERCENTAGETOEIGHT‐YEARTOTAL(2009‐2016)13
FIGURE11.VFRSDATA:TOTALNUMBEROFINCIDENTSPERMONTH(2009TO2016)14
FIGURE12.VFRSDATA:TOTALVEHICLERESCUESPERMONTH(2009TO2016) 14
FIGURE13.VFRSDATA:TOTALFIREINCIDENTSPERMONTH(2009TO2016) 15
FIGURE14.VFRSDATA:DISTRIBUTIONOFTIMEBETWEENSUCCESSIVECALLS,INMINUTES 15
FIGURE15.VFRSDATA:RESPONSETIMEDISTRIBUTION(ALLINCIDENTTYPES) 16
FIGURE16.VFRSDATA:ONSCENETIMEDISTRIBUTION(ALLINCIDENTTYPES) 17
FIGURE17.VFRSDATA:SPATIALDISTRIBUTIONANALYSISFORSPECIFICINCIDENTTYPES(2009‐2016)18
FIGURE18.VFRSDATA:ESTIMATEDAPPTOTDISTRIBUTION(ALLINCIDENTTYPES) 19
FIGURE19.INTEGRATIONOFEMPIRICALDATAFORSIMULATIONOFTHEINCIDENTLIST 20
FIGURE20.INCIDENTGENERATIONENGINE:MAINCOMPONENTS 21
FIGURE21.AGENTSUSEDINTHEMODEL 23
FIGURE22.SUBDIVISIONOFTHECITYOFVAUGHANINTOFIREDISTRICTSANDCURRENTFIRESTATIONLOCATIONS 24
FIGURE23.MODELHOMEPAGEANDSETTINGPARAMETERS 28
FIGURE24.STATIONSTATUSANDRELATEDSYMBOLS 28
FIGURE25.AGENTS’STATECHARTS 30
FIGURE26.PERFORMANCEINDICATORSVISUALIZEDDURINGSIMULATION 31
FIGURE27.DISTRIBUTIONOFSIMULATEDRESPONSETIMES,JANUARY 32
FIGURE28.DISTRIBUTIONOFSIMULATEDRESPONSETIMES,FEBRUARY 33
FIGURE29.DISTRIBUTIONOFSIMULATEDRESPONSETIMES,MARCH 33
43
FIGURE30.DISTRIBUTIONOFSIMULATEDRESPONSETIMES,APRIL 34FIGURE31.DISTRIBUTIONOFSIMULATEDRESPONSETIMES,MAY 34FIGURE32.DISTRIBUTIONOFSIMULATEDRESPONSETIMES,JUNE 35FIGURE33.DISTRIBUTIONOFSIMULATEDRESPONSETIMES,JANUARY–JUNE 35FIGURE34.AVERAGETRAVELDISTANCEPERINCIDENT(INMETERS)FORFIRSTRESPONDERVEHICLES
(180REPLICATIONS) 37
FIGURE35.SIMULATEDNUMBEROFINCIDENTS,BYREGION(180REPLICATIONS) 37FIGURE36.RESPONDINGVEHICLE/CREWOVERALLTIMEBREAKDOWN,BYREGION(180REPLICATIONS)38
44