2019 MCM/ICM Summary Sheet: "校苑数模"微信公众号，获取更多资料 math-o 获取免费课程

请关注“校苑数模”微信公众号，获取更多资料添加微信 math-o 获取免费课程
Team Control Number

For office use only For office use only
T1
T2
T3
1908286 F1
F2
F3
Problem Chosen
T4 F4
B
2019 MCM/ICM Summary Sheet
Hurricane Maria Damage left Puerto Rico people a lot of pain and devastation. HELP,Inc., an NGO,
is attempting to improve its response capabilities by designing a transportable disaster response system
called "DroneGo".
In order to better assess the transport capability of the drone fleet, we establish a Transport Capability
Evaluation Model of the drone fleet, so that we can maximize the medicine transportation capability of
the drone fleet when the disaster location is unknown. In this model, we use the algorithm of linear
programming to solve the limitations of volume, weight and other simple factors. After that, we used
公模
an algorithm based on three-space segmentation and improved Monte Carlo simulation to solve the size
号
limitation of the objects. As a result, we obtain a reliable evaluation of the transport capability of drones,
and it can be found that among the drones, the transportation capability of drone G is the strongest.
Through the research on the reconnaissance method of the drones, we get the Reconnaissance Ca-
信数
pability Model of the drone fleet. In this model, we find that the reconnaissance capability of drone
众
B is the strongest. Based on these two models (TCM & RCM) and abundant geographic information,
we developed a Geospatial Analysis Model. We employ the Analytical Hierarchy Process (AHP) in the
space of all position, obtaining a spatial distribution of drone fleetaŕs
˛ transportation capability and re-
connaissance capability. Then, we establish an overall Efficiency Evaluation Model for the drone fleet,
微苑
using a nonlinear programming with a variable parameter to obtain the optimal solution of the drone
fleet overall efficiency and its spatial location. Consequently, we can determine a configuration of the
drone fleet and each container’s drop point. DroneGo users can also adjust the variable parameter de-
pending on whether they care more about their fleetaŕs ˛ transport capability or reconnaissance capability
校
to determine different options. For the problem of multiple containers, we use a dynamic programming
rule that allows users to arrange multiple containers.
In considering that the transportation capability of the drone fleet is as important as the reconnais-
sance capability, we get results that the medicine delivery can be maintained for more than two months
and road reconnaissance coverage can reach nearly 60%. Meanwhile we construct an efficient Schedule
Arrangement Model and Route Arrangement Model of drones using an integer programming model
and an improved greedy algorithm, so that the drone fleet can be auto arranged.
The sensitivity analysis shows the strong robustness of our model. Meanwhile, we further discuss
the possibility of developing a DroneGo system software, and provide practicable advice to the HELP,
Inc. CEO.
Keywords: Multi-Objective Programming,Dynamic Programming, Drone Rescue Model, AHP
校苑数模收集整理，版权归原作者所有
Team请关注“校苑数模”微信公众号，获取更多资料
# 1908286 添加微信 math-o 获取免费课程
Page 1 of 22
Contents
1 Introduction 2
2 Assumptions and Justification 2
3 Transportation Capability Evaluation Model 3

3.1 Assumptions in This Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
3.2 Three Space Division Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
3.3 Putting MEDs into Bays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.4 Drone Transport Capability Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.5 Putting Drones and Bays into Cargo Conatainers . . . . . . . . . . . . . . . . . . . 7
3.6 Monte-Carlo Simulation And Results . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4 Geospatial Analysis Model 8

4.1 Drone Reconnaissance Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
4.2 Geographical Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4.3 Singal Location Determination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
4.4 Multiple Locations Determination . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
4.5 Results . . . . . . . . . . . . . . . .
公模 . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
5 Fleet Arrange Model 14
号
5.1 Schedule Arrangement Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
5.2 Drone Route Planning Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
信数
6 Sensitivity Analysis 19
众
7 Strengths and Weaknesses 19
7.1 Strengths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
微苑
7.2 Weaknesses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
8 Conclusion 20
校
Page 2 of 22
An optional Design of DroneGo
1 Introduction
People in disaster areas are in great need of timely emergency services. [1] Drones can help
a lot in emergency rescue. According to a drone manufacturer DJI [2], drones have saved 59
lives since 2013.
In this paper, we will
• Analyze three dimensional packing problem and develop Three Space Division Model
and improved Monte-Carlo Simulation to solve the problem of packing restrictions in the
transportation.
• Analyze the transportation capability and detective capability of each Drone type,apply
AHP method to considers the influence of many possible factors on the selection of con-
tainer drop location , including transportation, population, geography and medical infor-
mation.
• Build a model to determine drone payload packing configuration and solve it with a com-
公模
plex integer programming quickly and accurately.
号
• Improve greedy algorithm and figure out the drone route planing quickly and accurately.
信数
2 Assumptions and Justification
众
We make some general assumptions to simplify our model. These assumptions together
微苑
with corresponding justification are listed below:

• There are no charging devices for drones. In 2017, Hurricane Maria knocked down 80%
of Puerto Rico’s utility poles and all transmission lines. We can infer that electricity are
校
not available on the island. Our Cargo will not include charging equipment or generating
set, due to extra weight and inconvenience.
• The cargo weight does not influence the maximum flight distance of a drone. That
is, maximum flight distance is exactly the same with or without a cargo. We make the
assumption for simplicity. In reality, the weight of a delivery drone is often larger than
that of its cargo. It is quite reasonable to neglect the influence of cargo weight. And this
will greatly simply our model.
• For each Cargo Container, we will load one tethered Drone (type H). In natural disasters,
communication facilities are often affected to various degrees, and emergency communi-
cation becomes an urgent need. It can act as an emergency communication base station.
Tethered drones will greatly enhance the disaster response capability of DroneGo system.
• One drone can carry one drone cargo bay or no bay for every flight. If the drone carries
one, it can only deliver the bay to one location. Both of the two types of drone cargo
bay are "top loaded", that is, loaded on the top of a drone. If we distribute more than one
bay at a flight, connections between cargo bays could be unstable. We can deliver a drone
cargo bay to only one place since it is an integral package that cannot be open in transit.
• Helicopters can transport cargo containers to any given place in the disaster area. In
disaster-affected areas, timely external assistance is urgently needed to restore order. There-
fore, we consider air transportation rather than sea transportation. Although Puerto Rico
is an island, air transportation will bring adaptability to our model in other situations.
• Communication and data transmission of drones are always available. We can achieve
telecommunication by radio communication or satellite communication. Drone commu-
nication dose not rely on ground devices.
Page 3 of 22
• For the ith type of drone, its maximum Flight Distance, denoted by Disti is in direct
proportion to the product of Speed and F lightT ime(N o Cargo). From that assumption,
we have Disti = α × Speed × F light T ime (N o Cargo), where α represents distance-loss
coefficient. Here we neglect the effect of weather (especially wind) and different battery
and motor of different drones of a specific type. To determine α, we note that there are
mostly mountains with few plains. And Puerto Rico has a mean elevation of 261 m. [3] The
high elevation shortens the maximum flight distance. We have α < 1. For simplification,
we choose α = 0.9.
More detailed assumptions will be listed if needed.
3 Transportation Capability Evaluation Model
In this section, we will evaluate transportation ability of each type of drones. For simplicity,
we will focus on the need of medical transportation here. We will incorporate road reconnais-
sance into our model in Section 5.
3.1
公模
Assumptions in This Model
号
We will use the following assumptions in this section:
信数
• All Cargo Containers(CC), Drone (in shipping containers), Drone Cargo Bays (DCB),
众
and Medical packages (MP) are cuboids.
• Drones and Drone Cargo Bays should be placed into Cargo Containers in parallel,
which is "parallel placement". That is, the inner subface of Drones or Bays must be par-
微苑
allel to the corresponding surface of the Cargo Container. There are two reasons. For one
thing, the parallel placement is one of the most commonly used packing methods. For
another thing, cargos can be easily placed in this way and have a high stability.
• The thickness of Drone Cargo Bays can be neglected.
校
• Goods can be arbitrarily rotated, but must still satisfy the "parallel placement" condi-
tion.
• As is interpreted in the problem description, each Cargo Container’s contents should
be packed, so all Medical Packages should be packed in Drone Cargo Bays. We do not
consider Medical Packages directly placed in a Cargo Container.
3.2 Three Space Division Model
3.2.1 One Goods Condition
To design a packing configuration for DroneGo, we need to put MEDs into Bays and put
Bays or Drones into Containers. Due to various kinds of goods in the problem, this is a three-
dimensional heterogeneous container loading problem. We use the Three Space Division
Method and Monte Carlo Simulation to solve it.
Note that we refer to a rectangular container that contains goods as a box. As is shown in
Figure 1, L, W and H is the length, width and height of the container, respectively. Similarly,
l, w and h is the length, width and height of the cargo, respectively. We set up a Cartesian
coordinate at the left bottom vertex of the container. Note that we consider the position of the
container is unchanged. We put the cargo into the container, so that two left bottom vertexes
coincide. Arrangement problem is a way to think about ways to put the cargo. To put l, w and
Page 4 of 22
h into x, y and z coordinates, we have A33 = 6. There are six ways to put the cargo inside. That
is, there are six possible combinations of the coordinates of the diagonal vertex of the cargo.
Figure 1: Three Space Division
For every specific combination, we can see that the rest space in the container are cut into
three parts (spaceX, spaceY and spaceZ) , see figure 1. One possible sequence (shown in Figure
1) to generate the three parts is
1. Procedure X: Cut the remaining space using a plane perpendicular to the X-axis and get
spaceX.
2. Procedure Y: Cut the remaining space using a plane perpendicular to the Y-axis and get
公模
spaceY.
号
3. Procedure Z: Cut the remaining space using a plane perpendicular to the Z-axis and get
spaceZ.
We refer to this sequence as a X-Y-Z sequence. The rest are similar. Arrangement problem
信数
is a way to think about the space cutting process. To generate a sequence from procedure X, Y
and Z we have A33 = 6.
众
Thus, if a cargo can be put into the cargo, we have the total cases of three space division:
微苑
6 × 6 = 36 Then we consider every new space as a new container. And the three space division
process goes on, until the cargo can not be put into any of the existing containers. In this way,
we can traverse all situations and find the optimal solution for a specific objective function.
校
3.2.2 Multiple Goods Condition
To clarify the problem, we will use the following notations:

• K = {1, 2, . . . , |K|} is the set of Boxes’ number;
• R = {1, 2, . . . , |R|} is the set of goods’ types;
• Mr = {1, 2, . . . , |Mr |} is the set of goods’ number (same type) ;
• sB = (L W H)T is Box Size column vector;
• sG T
r = (lr wr hr ) is the rth type of Goods Size column vector;
• Xk is the number of goods in the kth box;
•
1, if the mth Goods of Type r is in Box k
Ykrm = is an Existing Flag;
0, otherwise
• u = (xkrm ykrm zkrm )T is coordinates of the mth goods of Type r in Box k;
•
1, if the length of the mth Goods of Type r is parallel to the x-axis
lxrm =
0, otherwise
is a Configuration Variable;
•
lxrm lyrm lzrm
!
P= wxrm wyrm wzrm is Configuration Matrix.
hxrm hyrm hzrm
Page 5 of 22
Undefined elements in the matrix is defined similarly to lxrm .

In multiple-goods-condition, we consider the container loading problem as a Single-objective
integer programming problem. Constraint Conditions of the three-dimensional heterogeneous
container loading problem are listed below:
1. Boxes should contain all goods:
|K| |R|
X X
Xk = Mr
k=1 r=1
2. Corresponding Rule. Each goods should be placed into just only one box:
|K|
X
∀r ∈ R, ∀m ∈ Mr , Ykmr = 1
k=1
3. Configuration Constraint. Each edge in each goods should be parallel to just only one
edge of the corresponding box. Let P = (pab )3×3 , we have
3
X 3
X
pab = 1 pab = 1.
a=1 b=1
4. Box Size Constraint. All goods should be inside the corresponding box:
∀k ∈ K, ∀r ∈ R, ∀m ∈ Mr , Ykrm (u + P · sG
r ) ≤ s
B
We can solve the programming problem with a specific objective function.
3.3 Putting MEDs into Bays
Since there are only three types of MEDs and two types of Bays, putting MEDs into Bays
is relatively simple. In 3D heterogeneous container loading problems, space utilization(SU )
usually decreases as type of goods R increases. For simplicity, we only consider two types of
MEDs or less in a Bay. We do not need to consider an objective function in this process. Rather,
we list all possible combinations subject to its constraint conditions.
Let R = {1, 2, 3}, whose elements 1, 2 and 3 correspond to MED1, MED2 and MED3, re-
spectively. And |Mr | is the number of MEDr. For type q Bay (q = 1, 2), we have the following
constraint conditions:
 P|K|
Xk = 3r=1 Mr
P

 k=1
 |K|
P
P3k=1 Ykmr = 1, ∀r ∈ R, ∀m ∈ Mr


pab = 1 , (1)
 Pa=1
3
 b=1 pab = 1 G



Ykrm (u + P · sr ) ≤ sB
q , ∀k ∈ K, ∀r ∈ R, ∀m ∈ Mr
where sB T B T
1 = (8 10 14) , s2 = (24 20 20) .
For type i Drone, an additional constraint is that total weight cannot exceed Max Payload
Capability(M P Ci ):
2|M1 | + 2|M2 | + 3|M3 | ≤ M P Ci (2)
(1 and (2) determine whether we can put a set of MEDs into a Bay. If that set of MEDs works,
we calculate its space utilization:
P3
(|Mr |lr wr hr )
SU = r=1 , (3)
Lq Wq Hq
Page 6 of 22
where (l1 w1 h1 )T = (14 7 5)T , (l2 w2 h2 )T = (5 8 5)T , (l3 w3 h3 )T = (12 7 4)T , (L1 W1 H1 )T =
(8 10 14)T , (L2 W2 H2 )T = (24 20 20)T .
We draw the space utilization plot of Drone D (Bay 1) and Drone F (Bay 2), see Figure 2.
From the figure we can see that space utilization of Bay 2 is relatively low. The reason is that
the inequality (2) is strict. In most cases, the max payload capability is so small that the all sets
of MEDs that satisfy (2) can be loaded. So the sub-figures for Bay 2 clearly shows the inequality
(2).
Figure 2: Space Utilization with different MEDs and different Bays
From Figure 2, we conclude that the Transport Capability matrix is

1 1 1
 
 2 4 2 
 7 7 4 
 
TC =   2 4 2 .
 (4)
 7 7 5 
11 11 7
 
10 10 6
3.4 Drone Transport Capability Model
Since we only consider medicine transport in this section, there must be a corresponding
Drone Cargo Bay on each drone. The number of Typei Drones to carry N umM EDj MEDjs is
N umij = N umM EDj /T Cij
. Note that we use N umij to evaluate transport capability, so it may be a decimal. The volume
needed when one Typei Drone carries all MEDs for one day is
X3
ViM ed Day =( N umij )(Vidrone + ViBay )
j=1
. The number of days that a Typei Drone can continue transporting MEDs to all needed place is
Daysi = Vc ontainer/ViM ed Day
Page 7 of 22
Transport Capability Index of Typei Drone is defined as the normalization of maximum day
Daysi :
T CIi = Daysi / max {Daysk }
1≤k≤7
We have two matrices:

2.97 0.0629
   
 15.92   0.3371 

 8.78 


 0.1859 

Days = 
 24.45 
 TCI = 
 0.5178 
 (5)
 41.46   0.8782 
28.98 0.6139
   
47.21 1.0000
3.5 Putting Drones and Bays into Cargo Conatainers
According to TCI matrix (5), Type E and G Drones have the best transport capability. We
will choose Type E and G as alternative transport drones. The corresponding Bays are Bays 2.
We will also load one Drone H (tethered drone) in each Cargo Container, because it helps to
build emergency communication.
Let R = {4, 5, 6, 7}. 4, 5, 6 and 7 represent Bay 2, Drone E, H and G, respectively. The number
of MEDs in Bay 2 only relies on Type i drone’s Max Payload Capability M P Ci . So we conclude
that for Drone fleet consisting of E and G, transport capability is only relevant to Total Load
Weight: X
T LW = M P Ci × |Mi | = 15|M5 | + 20|M7 | (6)
i=5,7
Our problem can be stated as
max T LW = 15|M5 | + 20|M7 |
Equation (??) shows that each Drone must carry one Bay, so the total number of Drones E
and G equal to the number of Bays 2. Equation (??) shows that each Cargo Container should
have one Type H Drone.
3.6 Monte-Carlo Simulation And Results
We use Monte-Carlo Simulation to get a satisfactory solution of the problem in Section 4.5.
Since this is a NP-completeness problem, it takes too long to traverse all situations. Our objec-
tive function will not have major changes with a narrow change of inputs. That is, the optimal
point is not an isolated singularity. That’s why we choose Monte-Carlo Simulation.
If the probability to find the optimal point is 10−5 for one simulation, the probability to find
the optimal point will be 1 − 0.999991000000 = 0.999954 with one billion simulations. Figure 3
is our simulation process using Three Space Division method. We start at a feasible solution
|M4 | = 50, |M5 | = 30, |M6 | = 1, |M7 | = 20.
Page 8 of 22
Figure 3: Monte-Carlo Simulation Process
Table 1: Satisfactory Packing Configuration for Each Container
Using the Three Space Division Model and Monte-Carlo Simulation, we obtain the satisfac-
tory packing configuration for each container (with or without Drone H). The result shows that
the ideal optimal solution is, at most, slightly better than our satisfied solution(space utiliza-
tion up to 97.97%). It also validates that Drone G has the best transportation capability, which
perfectly fit the highest Drone G TCI value.
4 Geospatial Analysis Model
4.1 Drone Reconnaissance Model
For a video-capable drone, in order to effectively estimate its maximum viewing radius,
rview , we make the following assumptions:
1. The height between a drone and the ground is 20 meters.
2. The drone’s lens can swing back and forth from side to side, with unlimited viewing
angles.
3. The camera on a drone (if it has one) is a two megapixel camera (a 1080P lens).
4. The resolution of road detection is one meter.
We see
tan(θ + ∆θ) = rview /20
, (7)
tan θ = (rview − 1)/20
where
Page 9 of 22
• θ is the camera’s forward viewing angle;

• ∆θ is the smallest resolution angle;
• rview is the maximum viewing radius.
For a two megapixel camera, we have ∆θ = 3 × 10−3 rad. Plugging it into (7), we obtain

θ = 1.322 rad
.
rview = 79.7 m
Since rview is close to 100 m, we can divide the island map into grid on the order of one
hundred meters, as is shown in Figure[ÍijÆň]. Every grid covers a 100 m × 100 m square. Cor-
responding to the longitude and latitude,(x, y) is the grid point coordinates of the position.
4.2 Geographical Features
We divide the location selection problem into two major parts Transport and Reconnais-
sance. Then we decide that the information of highway, population, geographic and medical
are important to Transport and Reconnaissance. Using the map in the attachment and from
Puerto Rico Maps [4], we find seven features for the four types of information, see Table 2.
Table 2: Features (F O is the Feather Order)
FO 1 2 3 4 5 6 7
Hospital
High
Main Highways Population location
feature(FO) population Harbours Altitude
highways & roads distribution & medical
areas
needs
For a certain position, Feature1 means the main highways. Feature 2 means the highways
we extract from attachment 1. Feature 3 means the density of population [4]. Feature4 means
high population areas from the given map. Feature 5 means the distance from harbors. [3]
Feature 6 means the altitude. Feature 7 means the index considering transporting MEDs.
The feature maps are shown in Figure 4. Note that all feature maps are normalized, so that
for every map, the sum of all values is 1 :
XX
F eature(F O, x, y), ∀F O ∈ {Z|1 ≤ F O ≤ 7},
x y
where F eature(F O, x, y) is the geographical distribution function of f eatureF O .

The weight of hospital j
3
X
HOSj = M edW ighti × M edN umberji
i=1
We assign the values in hospital map.

We calculate Transport cost for MEDj by Demand number for MEDj:
3
X
M edW eightj = M edj / M edi
i=1
Page 10 of 22
We have
MedWeight = (0.2949 0.2681 0.4369)T j = 1, 2, 3 (8)
From (8) we see M edW eight2 is smallest, because MED 2 has the least demand.
4.3 Singal Location Determination
4.3.1 Detail Information of TI and RI
T Idetail is an evaluation index for Drone Type i of feature F O in position p1 (x, y) and
evaluates the detailed transport capability. Let p1 be the target position for Cargo Container.
Let p2 be the corresponding point in feature map, we have
X
T Idetail (i, F O, p1) = T Iα (i, p1 , p2 )F eature(F O, p2 ).
p2
P
Similarly, RIdetail (i, F O, p1) = p2 RIα (i, p1 , p2 )F eature(F O, p2 ).
T Iα is a function of i, p1 and p2 :
T Iα (i, p1 , p2 ) = T Idistance (i, p1 , p2 )T Icover (i)T CI(i)
Similarly, RI α (i, p1 , p2 ) = RIdistance (i, p1 , p2 )RIcover (i)RCI(i)

T Idistance measures Transporting Index with flight distance:
dis(p1 ,p2 )

1− βT I , dis < DIS
T Idistance = DIS(i) , (9)
0, otherwise
where
• dis(p1 , p2 ) is the distance between p1 , p2 ;
• DIS(i) is the longest flight distance.
DIS(i) = α × Speed × F light T ime (N o Cargo), see(2) (10)
The function (9) shows that closer points are more important within coverage area. βT I = 0.2
produces a linear transformation.
Similarly,
dis(p1 ,p2 )

1 − (1 − )βRI , dis < DIS
RIdistance = DIS(i) , (11)
0, otherwise
where βRI = 0.2. βRI produces a linear transformation. Compare (9) and (11), function (11)
shows that in reconnaissance, farther points are preferred, because the close ones can be detect-
ed on the way to medical
T Icover measures Transporting Index with coverage area. The effectiveness of each point
in coverage area is equally important.
T Icover (i) = 1/(π · (DIS(i))2 · βT0 I ), (12)
where βT0 I is determined by βT I ; βT0 I = 13/15.

0 0
Similarly, RIcover (i) = 1/(π · (DIS(i))2 · βRI ), βRI = 14/15
Page 11 of 22
T CI measures Transporting Index with and Transporting Capability of Drone Type i with
unit volume. We computed the result of T CI(i) in matrix TCI (5). Similarly, We calculate
RCI1 (i) = (DIS(i))2 /V (i),where V(i) is the total volume of Type i drones and corresponding
Bays. Then we normalize it as RCI(i) = RCI1 (i)/ max RCI1 (p)
1≤p≤7
After we calculate T Idetail (i, F O, x, y) and RIdetail (i, F O, x, y), we should take all features into
account and calculate the integrate Transport Index of all features T I(i, x, y):
X
T I(i, x, y) = wT I (F O) · T Idetail (i, F O, x, y) (13)
FO
P
Similarly, RI(i, x, y) = FO wRI (F O) · RIdetail (i, F O, x, y). We obtain a Reconnaissance Capa-
bility Index matrix
0.0768
 
 1.0000 
 0.1105 
 
RCI = 
 0.1850 
 (14)
 0.1190 
0.1782
 
0.1194
Matrix (14) shows that Drone B is the best type to detect highways and roads. Note that Drone
B is a lot better than the other types.
We listed a few results of the detail information of TI and RI based on Puerto Rico case (see
Figure 4).
(a) Main Highways (b) Highways and Roads (c) Populated Places
(d) TI Drone G Main Highways (e) TI Drone G Highways & Roads (f) TI Drone G Populated Places
(g) RI Drone E Main Highways (h) RI Drone E Highways &c Roads (i) RI Drone E Populated Places
Figure 4: Samples of Detail Information of TI and RI
4.3.2 TI and RI
To determine the weight of different features, we employ Analytic Hierarchy Process (AHP).
Since those features in Table 2 have different effects on TI and RI, we use AHP twice for TI and
RI. The comparison matrices are listed below:
Page 12 of 22
(a) Comparison Matrix For TI (b) Comparison Matrix For RI
To test the consistency of the matrices, we calculate the Consistency Ratio (CR), which is
defined as the ratio of Consistency Index (CI) to Average Random Consistency Index (RI 0 ).
n = 4 , RI 0 = 0.90, CR = (λmax − n)/(n − 1). For TI, CI = 0.0532, CR = 0.0598 < 0.1. For RI,
CI = 0.0526, CR = 0.0591 < 0.1.so the consistency of the matrices is confirmed.
As is indicated in Figure 6, for Transporting Index, Medical information is the most im-
portant factor weighting 67.03%. That is because hospital destination of a Drone transporting
MEDs. For Reconnaissance Index, highway information take up the highest at 60.11%. One
possible reason is that the flight distance of Drones is limited. Drones must get close enough to
detect highways.
Then we put the weight terms into (13) and (4.3.1), and get T I and RI (see Figure5).
(c) TI Drone B (d) TI Drone E (e) TI Drone G
(f) RI Drone B (g) RI Drone E (h) RI Drone G
Figure 5: Samples of TI and RI
Figure 6: Sections and their weight when determining location using AHP
Page 13 of 22
4.3.3 Efficiency Evaluation
To show the relative relationship between TI and RI, we introduce an important parameter
λ as the transport objective coefficient. 0 ≤ λ ≤ 1. λ shows the importance of transport. The
bigger λ is, the more important transport is.
The location selection problem is a multi-objective programming problem. We simplify it
to a single-objective-programming by using λ. Considering transport and reconnaissance, our
objective function is the overall efficiency:
λ 1−λ
E = T IALL RIALL , (15)
In function (15),we define the Transport Index for all types of drones:
7
X
T IALL = v(i) · T I(i), (16)
i=1
where v(i) is the volume percentage of Type i drones and corresponding Bays in all types. Since
v(i) = V (i)/ 7p=1 V (p), we have 7i=1 v(i) = 1.
P P
P7
Similarly, RIALL is the RI for all types of drones: RIALL = i=1 v(i) · RI(i)
We compute the Best Efficiency for each position (x, y). Our programming problem can be
stated as
λ 1−λ
max E = T IALL RIALL
X7
s.t. v(i) = 1
i=1
v(i) ≥ 0, ∀i ∈ {Z|1 ≤ i ≤ 7}
(a) lambda = 0.2 (b) lambda = 0.4 (c) lambda = 0.5
(d) lambda = 0.6 (e) lambda = 0.8 (f) lambda = 1.0

Figure 7: Single Location Determination Based on Different λ Values
We apply this method to the Puerto Rico situation (see Figure 7).
4.4 Multiple Locations Determination
Selecting multiple best locations is a dynamic programming problem. Each location selec-
tion can effect the new selection. To show the effect,we set the two following rules:
Page 14 of 22
• Effects on Transportation. After each selection, we assume the Container satisfies 1/3
of all MEDs demands if its Drones and MED sullplies allow. Note that the container
should first provide MEDs to nearby medical centers. And we should renormalize the the
Medical Infromation feature.
• Effects on Reconnaissance. After each selection, we assume Drones of the Container
can detect all highways within the container’s Maximun Reconnaissance Scope. High-
way Information feature within that scope should decrease. We will multiply them by a
coefficient of r0 = 0.1.
4.5 Results
Table 3: The Result of Multiple Locations Determination Modle(λ = 0.5)
Table 4: The Location Selection Related To λ
5 Fleet Arrange Model
5.1 Schedule Arrangement Model
5.1.1 Satisfactory Solution on MEDs
We maximize the ability of DroneGo fleet to satisfy the anticipated Medical Package De-
mand. According to the choices of drone type given in Part B, we will only employ drone B, E
and G to transport MEDs. First, we search satisfactory solution on how to put MEDs into a Bay.
Then we can use those schemes to assemble a best schedule so that we maximize the ability of
DroneGo fleet to satisfy the anticipated Medical Package Demand.
We will use the following notations in this section:
• M EDjweight is the weight of MED j;

Page 15 of 22
• M EDjN um is the number of MED j put into a bay in a satisfactory solution;

• Drone 1,2 and 3 correspond to B,E and G, respectively;
• kij is the jth scheme for Type i drones;
• Ki = (ki1 ki2 . . . kir )T is the set of all the satisfactory schemes for drone i.
By using the Transport Capability matrix (4) in Section 4.3, we demonstrate that for drone B,
E and G, only Max Payload Capacity (MPC) will play a decisive role on whether a scheme is fea-
sible. Actually, satisfactory solutions must be at the feasible region. Consequently, satisfactory
solution for a Type i drone should subject to the stated constraints:
3
X
M EDweight
j × M EDnum
j ≤ M P Ci
1
3
X
M EDweight
j ×M EDnum
j + min M EDweight
j > M P Ci
j
1
We use integer programming method to search feasible solutions.

ki1 is the 1st solution.
Eventually we get K1,K2 K3. for drone 1 2 3 respectively has 3,27,44 satisfactory schemes.
5.1.2 Maximum Effective Days Model
Now we have known all satisfactory schemes on how to put MEDs into a Bay for drone B,E
and G.To evaluate the the ability of DroneGo fleet to transport Emergency Medical Package,we
make an assumption: Every delivery location’s need for MED should be satisfied equally.
Since delivery location has the certain anticipated Medical Package Demand everyday,we
can calculate a Maximum supply days if the type and number of drone and the schemes for
each corresponding drone are determined.
we will use the following notations:
• M EDsneed = (mj1 mj2 mj3 ) is the need of sth delivery location within the range of trans-
portation. mj1 , mj2 , mj3 is corresponding to the number of MED1,2,3.
• N umirs is the number of drone i which carry out the rth scheme ;
• |Ri | is the number of satisfactory schemes for drone i;
• S is the number of delivery location;
• Dayss is Maximum supply days for delivery position s.
Here is the Linear Programming model. For delivery position s,
| Ri |
3 X
X
max N umirs kis ≥ Dayss × M EDr
i=1 r
|S| | Ri |
XX
s.t. N umirs = Qk
s r
N umirs ≥ 0
Page 16 of 22
3 |P
P Ri |
N umirs kis
i=1 r
max Daysef f ective = min {Dayss } = min ( )
1≤s≤|S| 1≤s≤|S| MEDs
|S| | Ri |
X X
s.t. N umirs = Qk
s r
N umirs ≥ 0
Table 5: Optimal Drone Payload Packing Configurations And Drone Schedule
Through the use of integer linear programming, we can find the optimal drone payload
packing configurations by traversing all possible combinations of the satisfied scheme when
the number of each drone type and delivery locations’ MEDs need are certain. Now we apply
the Schedule Arrangement Model to the situation in Table 3. The result indicates that we can get
an optimal solution in a given situation. Moreover, the result directly gives the drone payload
packing configurations for every drone.
5.2 Drone Route Planning Model
Now that we have confirmed the drone delivery plan. Under this scheme, the basic delivery
strategy of each drone has been determined. Since we assume that each drone will not move
once it has delivered the goods, the extra distance of the drone other than delivery can be used
to detect the highways and roads. We can use a Greedy Search Algorithm to arrange individual
drones. The algorithm is shown in Figure 8:
Page 17 of 22
Figure 8: Individual Drone Route Planning Algorithm
• Searching nearby, that is, searching the nearest road. Since most roads are continuous, the
nearby search algorithm is simple, easy to operate and efficient.
• The searched road should not be searched any more, and it should be deleted from the
overall search targets.
• Delivery first strategy, because delivery is urgent, so if you choose between delivery and
search, then you must choose delivery.
In the algorithm, we call the hospital where the drones need to deliver as the target location.
The keys of the algorithm are listed as follows Searching nearby, that is, searching the nearest
road. Since most roads are continuous, the nearby search algorithm is simple, easy to operate
and efficient. The searched road should not be searched any more, and it should be deleted
from the overall search targets. Delivery first strategy, because delivery is urgent, so if you
choose between delivery and search, then you must choose delivery.
For the drone fleet as a whole, we need to select different drones for reconnaissance. Since
short-range drones cannot detect points far from the targets and containers, we should priori-
tize the path of these drones to maximize their reconnaissance potential. The number of drugs
required by the hospital will be changed each time the drone delivers the goods, and the next
delivery mission of the drone will also be changed accordingly. More specifically, if the delivery
demand of a certain drug in a hospital has been met, the drone loaded with that drug should
not be used as the target place.
Further, to improve the algorithm, we can find that among the three selected drones, drone
B, E and G, the transport capacity of drone B is the weakest, but its reconnaissance capability
is the strongest. Even if we deduct the medicine delivery amount of drone B, the effect on
the overall drug delivery amount is not significant. Therefore, we should use drone B only
for reconnaissance. If we apply the above strategy to the 3 container transport model with
λ == 0.5, we can obtain the results of transportation path in Figure 10.
Page 18 of 22
Figure 9: Drone Fleet Route Planning Algorithm
Under the model of λ = 0.5, our road reconnaissance coverage for Puerto Rico reached
59.86% (see Figure 10). We can find that, in order to meet the needs of the three hospitals n-
earby Puerto Ricoaŕ˛ capital, we use a lot of drones to transport medicine, so the reconnaissance
coverage rate of this area is not high, however, in the area of transport demand is relatively
small, the container 3, our reconnaissance coverage is very high. This result is consistent with
our hypothesis that λ = 0.5, which means reconnaissance and transportation is equally im-
portant. At the same time, since we consider the range reduction of drone affected by various
reasons, that is, the ratio between the total range used and the ideal total range calculated is
α = 0.9, the reconnaissance capability of drone fleet may be better in actual use.
(a) Container 1(C1) (b) C1 drone B (c) C1 drone G
(d) Container 3(C3) (e) C3 drone B (f) C3 drone E
(g) Container 2(C2) (h) C2 drone E (i) All(routes under reconnaissance

in black, undetectcted in gray)
Figure 10: Drone Routes(the yellow line in the figure is the flight path of this drone, and the blue dot at
the end of the yellow line represents the final position of the drone)
Page 19 of 22
6 Sensitivity Analysis
As is shown on Geospatial Analysis Model, we use AHP to get the weight vector of wT I
and the weight vector of wRI . However, AHP is based on subjective inputs to some extent.
Aiming to determine how robust our rankings are in the AHP process, we use the following
proportionality equation for adjusting weights:
1 − wp0
wj0 = wj ,
1 − wp
where wj0 is the new weight, wp the original largest weight of the criterion to be adjusted,and wp0
the value after the criterion is adjusted.
Table 6: Sensitivity Analysis Result of AHP
We respectively apply a −10%, −20%, +10%, +20% variation to the wT I and wRI ,and record
the the results under each circumstance, respectively. The following is the new results:
In the table above, note that Latitude and longitude are used to indicate location and dis-
tance between original and new position is shown if position changes. We can conclude that
• Above all, the table shows the strong robustness of our model.It is conspicuous that the
results only change slightly,notwithstanding we let wDI or wTI have a sizable change.
• Remarkably, some positions of container and drone proportion donaŕt
˛ vary with different
wT I and wRI . Via analyzing the corresponding data, a reasonable explanation is that this
position or proportion has much better performance than those adjacent choices.
7 Strengths and Weaknesses
7.1 Strengths
• A model based on Three Space Division Model and improved Monte-Carlo Simulation
Page 20 of 22
can solve the problem of packing restrictions in the transportation of goods and have
high realizability by considering weight and volume limitations.
• Based on the AHP method,the Geospatial Analysis Model fully considers the influence of
many possible factors on the selection of container drop location , including transporta-
tion, population, geography and medical information. Therefore, the result of the model
is highly credible.
• Complex integer programming model and improved greedy algorithm is able to give the
drone schedule planning and route planning quickly and accurately.
• Flexibility and adaptability are also advantages of the Location Determination Model.
When the company change its strategy, new decisions can be quickly derived without
modifying the algorithm.
7.2 Weaknesses
• While utilizing improved Monte-Carlo Simulation,the quality of the result may not achieve
our high expectation.
• The rule of how to transport MEDs to different delivery position is uncertain.It can be
further improved to meet different actual needs
• The correctness of our conclusions is related to the accuracy of the data collected and the
estimates of transportation, population, geography and medical information.
8 Conclusion
To help with post-disaster relief, we designed a prototype of the DroneGo. The system estab-
lishes the Drone Transport Model and Drone Survey Model, taking into account the characteris-
tics of drone, medicine and container such as weight and size, as well as a variety of geographic
information such as road information, population information, rescue information and medical
information. We conclude that the Transportation Capability of Drone G is the strongest and
the Reconnaissance Capability of drone B is the strongest. When users considered medicine
delivery to be as important as road reconnaissance, we got results that the medicine delivery
can be maintained for more than two months and road reconnaissance coverage reached nearly
60%.
To further provide advice to the DroneGo users, in our system, users can also input a d-
ifferent transport/reconnaissance parameter, i.e., different emphasis on transport and recon-
naissance. In this case, our DroneGo system can provide different fleet configurations and
container drop points to meet their needs. Meanwhile we construct an efficient schedule ar-
rangement and route arrangement model of drones using an integer programming model and
an improved greedy algorithm, so the drone fleet can be auto arranged.
References
[1] Wikipedia, “Hurricane maria,” https://en.wikipedia.org/wiki/Hurricane_Maria#Puerto_

Rico_3, accessed January 29, 2019.
[2] M. Winn, “We need to be using drones to rescue harvey victims,” https://www.cia.gov/
library/publications/the-world-factbook/geos/rq.html#Geo, accessed January 29, 2019.
[3] CIA, “Central america :: Puerto rico,” https://w3techs.com/technologies/overview/
content_language/all, accessed January 29, 2019.
Page 21 of 22
[4] “Puerto rico maps,” http://www.topuertorico.org/maps.shtml, accessed January 29, 2019.
Memo to CEO of HELP, Inc.
To: HELP, Inc. CEO

From: Team 1908286
Date: January 28, 2019
Subject: Study of DroneGo disaster response system based on Puerto Rico hurricane in 2017
In the post-disaster relief, unmanned aerial vehicles, also called drones, are not limited by
road conditions, so they can quickly provide medicines and other urgent materials to the af-
fected areas, which can alleviate the shortage of local post-disaster medical drugs. Meanwhile,
drones can also be used to survey the highways and roads situation and support post-disaster
relief. We aim to develop a "DroneGo" disaster response system based on a drone fleet, which
can be used to provide medical assistance to the people in the disaster area, as well as to provide
information on the disaster situation to the rescue workers.
In 2017, the worst hurricane to ever hit the United States territory of Puerto Rico left the
island with severe damage. The DroneGo system needs to be designed to better help NGOs
use drones for post-disaster exploration and drug delivery missions in similar disasters. The
system can reasonably plan the selection of a certain drone fleet, container delivery location,
drone medicine delivery schedule and route planning of exploration road based on the known
properties of drone, medicine and containers, as well as, medical needs and highways and
roads reconnaissance needs in disaster areas and other relevant information.
We designed a prototype of the DroneGo system based on official drone, medicine and con-
tainer information and applied it to the disaster situation in Puerto Rico. When users con-
sidered medicine delivery to be as important as road reconnaissance, we got results that the
medicine delivery can be maintained for more than two months and road reconnaissance cov-
erage reached nearly 60%. Users can also input a different transport/reconnaissance parameter,
i.e., different emphasis on transport and reconnaissance. At the same time, our DroneGo sys-
tem can also quickly obtain the program of drone fleet composition, delivery and arrangement.
In addition, our DroneGo system can also form a model for drone transportation and recon-
naissance evaluation for more geographic information (such as population distribution, road
information, rescue information, medical information, etc.). The approach we are using can be
further refined and extended to develop a post-disaster rescue drone fleet arrangement system.
According to our ařDroneGo
˛ aś
˛ system, we can develop a software (see Figure 11), which will
greatly help post-disaster relief missions.
Firstly, for the loading problem of drone fleet, we use an indicator of drone fleet transport
capacity to form our drone Transport Capacity Model to solve the problem of maximizing the
drone fleet medicine transport capacity when the disaster location is unknown. Among them,
the biggest obstacle we encountered is the limitation of goods packing. We developed an al-
gorithm based on the three-space division method and an improved Monte Carlo simulation,
which can better solve this problem in a short time. At the same time, we also consider the
weight, volume and other restrictions, so that our results are highly achievable. In terms of
specific solutions, Drone G is considered to be the most transportable.
Secondly, if we got the information of the disaster area, we can use our DroneGo system
based on spatial geographic information, which integrated the previous drone Transportation
Capability Model and a drone Reconnaissance Capability Model. Combining with different ge-
ographic information, we used Analytic Hierarchy Process (AHP) to analyze the distribution
Page 22 of 22
of drone transportation capability and reconnaissance capability along with the location. In or-
der to better balance ařDroneGo
˛ aś
˛ system users to transportation capacity and reconnaissance
ability of different needs, we set up an overall Efficiency Evaluation Model, using the nonlinear
programming and dynamic programming to get the specific configuration of the drone fleet
and container drop point selection. In the system, users can adjust the transportation / recon-
naissance parameter to get the response of different targets. Through the case study of Puerto
Rico, we are able to further verify the feasibility of our system.
We use an integer programming model and an improved greedy algorithm to construct
the Schedule Arrangement Model and Route Arrangement Model of drones. In drone fleet
and containers delivery plan chosen by our DroneGo system, we got the medicine transport
to maintain nearly two months, as well as the results of road reconnaissance coverage reached
nearly 60%. Finally, the sensitivity of AHP is analyzed to verify the robustness of the system.
If you want to know more details, please read our essay about this subject. We will be
glad to discuss more about DroneGo system with you in the future and follow through on any
decisions you make.
Attached: DroneGo system software frame, January 28 2019
Figure 11: DroneGo system software frame

2019 MCM/ICM Summary Sheet: "校苑数模"微信公众号，获取更多资料 math-o 获取免费课程

Uploaded by

Copyright:

Available Formats

2019 MCM/ICM Summary Sheet: "校苑数模"微信公众号，获取更多资料 math-o 获取免费课程

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

2019 MCM/ICM Summary Sheet: "校苑数模"微信公众号，获取更多资料 math-o 获取免费课程

Uploaded by

Copyright:

Available Formats

请关注“校苑数模”微信公众号，获取更多资料 添加微信 math-o 获取免费课程

Team Control Number

2019 MCM/ICM Summary Sheet

2 Assumptions and Justification 2

3 Transportation Capability Evaluation Model 3

4 Geospatial Analysis Model 8

5 Fleet Arrange Model 14

An optional Design of DroneGo

with corresponding justification are listed below:

3 Transportation Capability Evaluation Model

3.2 Three Space Division Model

3.2.1 One Goods Condition

Figure 1: Three Space Division

3.2.2 Multiple Goods Condition

To clarify the problem, we will use the following notations:

Undefined elements in the matrix is defined similarly to lxrm .

We can solve the programming problem with a specific objective function.

3.3 Putting MEDs into Bays

Figure 2: Space Utilization with different MEDs and different Bays

From Figure 2, we conclude that the Transport Capability matrix is

3.4 Drone Transport Capability Model

We have two matrices:

3.5 Putting Drones and Bays into Cargo Conatainers

Our problem can be stated as

max T LW = 15|M5 | + 20|M7 |

3.6 Monte-Carlo Simulation And Results

Figure 3: Monte-Carlo Simulation Process

Table 1: Satisfactory Packing Configuration for Each Container

4 Geospatial Analysis Model

4.1 Drone Reconnaissance Model

• θ is the camera’s forward viewing angle;

4.2 Geographical Features

Table 2: Features (F O is the Feather Order)

where F eature(F O, x, y) is the geographical distribution function of f eatureF O .

We assign the values in hospital map.

4.3 Singal Location Determination

4.3.1 Detail Information of TI and RI

T Iα (i, p1 , p2 ) = T Idistance (i, p1 , p2 )T Icover (i)T CI(i)

Similarly, RI α (i, p1 , p2 ) = RIdistance (i, p1 , p2 )RIcover (i)RCI(i)

DIS(i) = α × Speed × F light T ime (N o Cargo), see(2) (10)

T Icover (i) = 1/(π · (DIS(i))2 · βT0 I ), (12)

where βT0 I is determined by βT I ; βT0 I = 13/15.

Figure 4: Samples of Detail Information of TI and RI

(a) Comparison Matrix For TI (b) Comparison Matrix For RI

(c) TI Drone B (d) TI Drone E (e) TI Drone G

(f) RI Drone B (g) RI Drone E (h) RI Drone G

Figure 5: Samples of TI and RI

4.3.3 Efficiency Evaluation

(a) lambda = 0.2 (b) lambda = 0.4 (c) lambda = 0.5

(d) lambda = 0.6 (e) lambda = 0.8 (f) lambda = 1.0

4.4 Multiple Locations Determination

Table 3: The Result of Multiple Locations Determination Modle(λ = 0.5)

Table 4: The Location Selection Related To λ

5 Fleet Arrange Model

5.1 Schedule Arrangement Model

5.1.1 Satisfactory Solution on MEDs

• M EDjweight is the weight of MED j;

• M EDjN um is the number of MED j put into a bay in a satisfactory solution;

We use integer programming method to search feasible solutions.

请关注“校苑数模”微信公众号，获取更多资料添加微信 math-o 获取免费课程