Docstoc

A Novel Approach to Evacuation Route Planning

Document Sample
A Novel Approach to Evacuation Route Planning Powered By Docstoc
					                                  The goal of evacuationroute planning is to identify routes
                             that canbeusedto minimize the time requiredto move vulnerable
                             populationsto safe destinations.Evacuationroute planning is a
                             key componentof effective disasteremergencymanagement        and
                             homelanddefense   preparation. such,it is an importantchallenge
                                                           As
                             facing the U.S. Army, which must be ready and able to assistin
                             large-scale evacuations civilians. Sinceevacuationplanning is
                                                     of
                             computationintensive,high performancecomputing is essential
                             to meetingthis challenge.  Also, real-time solutionsare neededto
                             support evaluationof alternativescenarios(e.g. weatheq change
                             in transportationnetworks,or number of evacuees).
                                  Traditional warning systems convey only the threat
                             descriptions and the need for evacuation to the affected
                             population. Such systemsdo not consider capacityconstraintsof
      A Novel                the transportationnetworkand,asa result,haveproved inadequate
                             during actual evacuation events. When Hurricane Andrew was
                             approachingFlorida in 1992, evacuationannouncements to     led
                             tremendoustraffic congestion,general confusion, and chaos. In
  Approach to                very recent times, when Hurricane Rita approachedHouston,
                             evacuations  resultedin congestionon Texashighways for tens of
                             miles as seenin Fisure 1.
   Evacuation
                                                                            Figure l. Hurricane
Route Planning                                                              Rita evacueesfrom
                                                                            Houston clog I-45
                                                                            (FEMA.gov).




                                 It is now evident to planning authorities that to reduce
Shashi Shekhar QingsongLu,   congestionduring large-scaleevacuationsat all levels, effective
     and SanghoKim
                             evacuation route planners are needed that take into account
       AHPCRC-UM
                             the capacity constraintsof transportationnetworks. An overall
                             framework of evacuationroute planning is shown in Figure 2,
                             using the example of anothertype of disaster- a bio-chemical
                             attack. The base map and weather data are used as inputs in
                             the analysis of the plume dispersion of the bio-chemicals.
                             An evacuation route plan is generated from the analysis by
                             incorporating the demographic information and transportation




                                                           Anuv HPC Rtszancn CzwrsnBuusril{
                                                                                         Weather Data
        Base Map



                                            Plume Dispersion




                                                                                     Transportation Network
 Demographic Information

Figure 2. Bio-chemical Attack Analysis (images from www.fortune.com).


network. The evacuation route planning                 plan in real-time(e.g.seconds), edgetravel
                                                                                      (2)
                                                       time preservesthe FIFO    (First-In First-Out)
problem is formulated as follows:
     Given: ( 1)Atransportationnetworkwith             property, (3) edge travel time affects delays
non-negativeintegercapacityconstraints      on         at intersections,and (4) limited amount of
nodes and edges, (2) non-negativeinteger               computer memory.
travel time on edges,(3) the total number of                Previous methods were based on a
evacuees   and their initial locations, and (4)        linear programming [1] and suffered from
locations of evacuationdestinations.                   limitations. First, they did not scale up to
     Output: An evacuationplan consisting               large (e.g. >50,000 nodes) transportation
of a set of origin-destination routes and              networks in urban evacuation scenarios as
a scheduling of evacueeson each route'                  shown in Table 1, becausethey used time-
The scheduling of evacueeson each route
                                                                                                        s0.000
 should observe the capacity constraintsof              Number of Nodes     50         s00    5,000

the nodesand edgeson this route.                        Linear            0.1 mi n   z.) mm   108min    > 3 days
                                                        Programming
     Objective: Minimize the evacuation                 Method Running
 egresstime, which is the time elapsedfrom              Time

 the start of the evacuation until the last
 evacuee  reachesthe evacuationdestination'             Table 1. High computational complexity of existing
                                                        methodo linear Programming.
     Constraints: (1) Producean evacuation




                  CsNranBuusnu
   Anur HPC RESEARCH
 expandednetworks requiring large amounts                                          l *C C R P   *R el axl V l
 of computer storage and were aimed at
 computing optimal solutions that incurred            !       roco
                                                      o
 exorbitant computationalcosts.Second,they            g 6.     100

 required users to provide an estimateof the         :    6     10

 upper bound on the total evacuationtime, and     a:

 incorrect estimatesof the upper bound led to
                                                     o         0.1
 failure.                                         rd
                                                              !.0{
      To combattheselimitations, a novel geo-                             50            500             508t      500m
 spatial approach was developed, namely a         *CCRP               0.0s              0.69            9.38      108,'ts
                                                  +R€iaxlv            0.13              6.3S           189.51    6971.05
 CapacrtyConstrainedRoute Planner(CCRP)
 [2], which can be used to quickly identify     Figure 3. Run-time (log-scale)with respect to net-
 feasible evacuation plans. This approach       work size.
 provides an efficient decision support tool
 for homeland security officials to evaluate                                       CCRPa RelaxlV
 existing evacuation plans. It also allows       g''[oo
                                                 g
 them to determine plausible evacuation
plans for large transporlation networks in       ; 350
 an urban scenariowhen resourceconstraints       i: tco
                                                 si
                                                 trt
 or dynamic conditions make it infeasible or
uninterestingto find the optimal plan.
                                                 $rm
                                                 gI

      This approachhas two key ideas. First,     ; 200
                                                 (u
it models the capacity and occupancy of          3 150
                                                 at
each road segment as a time-series rather        ul t00
than fixed numbers becausethese attributes                           $0             590            $800         5s000
could changeover time during an evacuation.                                    N$mbor0f lrlodes{log$cale}
 Second, it repeatedly considers all pairs of
sources and destinations.In each iteration,     Figure 4. Quality of solution with respectto network
                                                size.
it schedules the evacuation of a group
of evacuees across the closest source-
destination pair. Special graph algorithms      algorithm over one of the existing linear
are used to eliminate redundantcomputation      programming approaches     named RelaxIV in
in this step. For smaller networks, where       a logarithmic scale.Even though the run-time
linear programming tools could be used,this     of CCRP showed remarkable improvement
approach produced high quality solutions        in terms of scalability, CCRP produced high
with evacuation times comparable to those       quality solutions (within 5 percent of the
achievedby linear programming methods.In        optimal evacuationtime) close to the optimal
addition, this approachsignificantly reduced    solution produced by RelaxIV, as shown
the computational cost by using much less       Figure 4.
computermemory.                                      Figure 5 shows the parallel formulation
     Evaluation included experiments on         of the CCRP algorithm. It parallelizesstep I
synthetic scenarios. Figure 3 shows the         of CCRP,which was a bottleneck step of the
performance superiority of the CCRP             algorithm. The parallel strategywas to divide




                                                                     AnmyHPC RESEARCH
                                                                                   Cturrn Buutruu
                                          Step I (N/P searches         Step I (NlP searches      Figure 5. Parallel formulation of
                                            on Processor l)              on Processor P)         CCRP.




     N shortestpath searches amongP processors,                                      as defined by Minnesota Homeland Security
     each processor thus having N/P searches.                                        and Emergency     Management. experiment
                                                                                                                      An
     Figure 6 shows the results of a parallel                                        was conducted    using the road network around
     version of CCRP for large problem sizes(50,                                     theevacuation   zoneprovidedby theMinnesota
     000 nodes).Speedup  improved as the number                                      Departmentof Transportation,     andthe Census
     of processorsincreased, with especially high                                    2000   population data for each affected city.
     speedupachieved by dynamic     parallel load                                    This case study showed that the approach
     balancing.                                                                      could be usedto improve existing evacuation
         Evaluation of these methods for                                             plans by providing higher capacitiesnear the
     evacuation planning included a real-world                                       destination and by choosing shorter routes.
     scenario around the Monticello nuclear                                          Due to the timelines of this casestudy,it was
                                                                                     invited for presentation at a congressional
                             -"r.*8J&    t-. O*wYlc...|.. GddM           I           breakfast[3] on GIS and Homeland Security.
                                                                                           The shortest path algorithm in this
                7                                                                    approachassumedthat the edge travel times
                6                                                                                                             It
                                                                                     includedtraffic delaysat intersections. also
          *     5
                                                                                     assumedthat the travel times were not time-
          !
           t{

          &.
                                                                                      dependent.Current plans are to incorporate
                                                                                      existing work in this area to addressthose
     :2
                                                                                      limitations.Another interestingpossibility for
     il
     I                                                                                future work is to integratethe CCRP approach
                nl
     :0                                                          81$
                                                                                      with the traffic assignment-simulation
                                        "*'",Y                                        approach to conduct stochastic simulation
     Figure 6. Parallel CCRP - experiment results.                                    of traffic. As the complexity and scale of
                                                                                      experimentsincrease,the computing power
     power plant near the Minneapolis/St. Paul                                        of the AHPCRC will play an important role
     metropolitanarea.The evacuationzonewas a                                         in the effort to develop scalable evacuation
     1O-mileradiusaroundthe nuclearpower plant                                        route planners. a


     References
     l. H.W. Hamacher and S.A. .;-andra.  Mathematical Modeling of Evacuation Problems: A State-of-the-Art
         Pedestrianand EvacuationDynamics, pages227-266,2002'
                                                                                                            A
     2. e. Lu, B. George, and S. Shekhar. Capacity Constrained Routing Algorithms for Evacuation Planning:
                                                                      Proc. of 9th
         Summary of Results,Advances in Spatial and Temporal Databases,            Int. SSTD '05, August 2005'
                    (http:i/www.cs.umn.edu/Research/shashi-group/paper_ps/evac-SSTD05.pdf).
     3.             UCC1S Congressional                            EvacuationPlanning for Homeland Defense( h t t p : / / w w w.
                                              Presentatio (02105104):
                                                        n
                    uc gi s.org/wi nter2004I Progt am.



l0                             Cryrzn Buuzrtu
                Anuv HPC RESEARCH

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:18
posted:3/17/2011
language:English
pages:4