computation

Article
Modification of the Bellman–Ford Algorithm for Finding the
Optimal Route in Multilayer Network Structures
Olga Timofeeva *, Alexey Sannikov, Maria Stepanenko and Tatiana Balashova

Department of Informatics and Control Systems, Nizhny Novgorod State Technical University, R.E. Alekseeva,
603950 Nizhny Novgorod, Russia; [email protected] (A.S.)
* Correspondence: [email protected]

Abstract: One of the actual tasks of the contemporary logistics business using the “just in time”
supply planning concept, is to distribute manufactured goods among the objects of the distribution
network in the most efficient manner at the lowest possible cost. The article is devoted to the
problem of finding the optimal path in network structures. The problem statement for multilayer
data transmission networks (MDTN), which is one of the possible representations of multimodal
transport networks, is considered. Thus, each MDTN layer can be represented as a separate type
of transport. The problem is solved by modifying the Bellman–Ford mathematical programming
algorithm. Load testing of the modified method was performed, and a comparative analysis was
given, including an assessment of speed and performance, proving the effectiveness of the results
of the study. Based on the results of comparative analysis, recommendations for using a modified
version of the Bellman–Ford algorithm for application in practical problems in optimizing logistics
networks are proposed. The results obtained can be used in practice not only in logistics networks
but also in the construction of smart energy networks, as well as in other subject areas that require
optimization of multilayer graph structures.

Keywords: optimization of logistics networks; multilayer data transmission network; modeling of

shortest route search on a graph; Bellman–Ford algorithm
Citation: Timofeeva, O.; Sannikov, A.;
Stepanenko, M.; Balashova, T.
Modification of the Bellman–Ford
Algorithm for Finding the Optimal 1. Introduction
Route in Multilayer Network
Nowadays, network structures are one of the most popular types of visualization of
Structures. Computation 2023, 11, 74.
various information in many areas of science and society. Graph visualization is widely
https://doi.org/10.3390/
computation11040074
used in the management of logistics and smart electrical networks, information security,
and many other areas of modern activities.
Academic Editors: Evelin Krmac and As a network structure, we will use a goods distribution network, which is a graph
Danijela Tuljak-Suban representation of the subjects that are necessary for transporting products from the manu-
Received: 27 January 2023 facturer to the end consumer. The aim of the work is to determine the optimal route for
Revised: 28 March 2023 delivering the product from the point of production to the final consumer, which provides
Accepted: 3 April 2023 for minimal costs for material, labor and time resources. Modeling the operation of such a
Published: 7 April 2023 network will optimize the plan for the supply of products, taking into account the impact
of external factors, which will ensure the rational work of the manufactory. Nowadays,
multimodal transportation, which is a more flexible type of transporting goods than classi-
cal commodity distribution networks due to the ability to use several types of transport,
Copyright: © 2023 by the authors. has become widely used in the logistics sector. To optimize such logistics networks, it is
Licensee MDPI, Basel, Switzerland. possible to use multilayer graph models.
This article is an open access article The main goal of supply chain management is to reduce economic costs and meet the
distributed under the terms and
needs of customers in the final product.
conditions of the Creative Commons
The supply chain is a set of links interconnected by information, cash and commodity
Attribution (CC BY) license (https://
flows and includes the stages of planning, procurement, production, delivery and return
creativecommons.org/licenses/by/
of goods.
4.0/).

Computation 2023, 11, 74. https://doi.org/10.3390/computation11040074 https://www.mdpi.com/journal/computation

Computation 2023, 11, 74 2 of 14

2. Literature Review
A goods distribution network, as one of the varieties of network structures, is a
connected structure consisting of system elements, including graph vertices and transi-
tion states between them. Such connections between vertices are referred to as edges.
Thus, in [1–4] the problem of supply chain management is solved with the optimization of
the parameters spent on the transportation of goods and profit maximization, respectively.
In [5,6], the problem of optimizing the consumption of electricity, ensuring its uninter-
rupted supply and restructuring the network in case of failure of the nodes of the graph
structure (switches) is solved. In this case, the nodes of the graph are network switches,
and the arcs are power transmission lines.
One of the main problems solved by network structures is the APSP problem (all
pairs shortest path problem)—the problem of finding the optimal route on a given space
of states. Over its almost half-century history, many algorithms have been created to
solve this problem, ranging from the method of exhaustive enumeration of all possible
solutions to modified genetic algorithms. Thus, in [7–9] a logistics network of multimodal
transportation is presented in the form of a graph structure. The nodes of the graph are
production points, warehouses and points of consumption of products, distribution centers
for loading and unloading goods and the arcs are various types of transportation. The
weight of the arc means the cost of transportation and includes various factors, such as
the time and cost of transportation, the seasonality of the product, its transportability,
and others.
Various deterministic, heuristic methods and artificial intelligence algorithms are
used to optimize network structures. One of the first methods was Dijkstra’s algorithm,
according to which each vertex of the graph was associated with the minimum known
distance from this vertex to the desired one, the so-called labels. Further, at each step, the
method “visits” the next vertex and tries to reduce the labels. The algorithm terminates
when all vertices have been visited. Currently, there are many new modifications of this
algorithm, considered in [10,11].
By now, Dijkstra’s algorithm has undergone different modifications and other algo-
rithms, based on it, have appeared. For example, the Floyd–Warshall algorithm gained
wide popularity. The method belongs to dynamic programming methods for finding the
minimum distances between graph vertices. The main advantage of the algorithm is that
it can be successfully applied to a weighted graph with both positive and negative edge
weights [12].
One of the latest algorithms developed for finding the shortest path in graph structures
was Jump Point Search (JPS), described by Australian scientists D. Harbor and A. Grad-
shtein in [13,14]. The algorithm was obtained by modifying the A* method [15] and applied
to an indefinite graph. The goal is to recursively traverse all points that can be reached by
an optimal path that does not pass through the current position. The exit from the recursion
is carried out after hitting the so-called “jump point successor” and the process for this
node starts again.
Thus, the problem of optimization of graph structures finds more application in prob-
lems of various applied areas. For this purpose, adaptation, modification and development
of previously known algorithms are carried out annually. In this article, a method for
solving the APSP problem for a network model consisting of several layers is considered.
The novelty of the work is the adaptation of the Bellman–Ford algorithm for finding the
shortest route in a multilayer data transmission network.

3. Problem Formulation
A multilayer data transmission network (MDTN) or a network model is a collection of
several layers representing a pool of undirected graphs Gm , m ∈ 1..z, where z is the number
m
of graphs in the model, built on a set of a vertices Vm ∈{v1 , . . . ,vn }. Edges of graph Gm Ei,j
m
have certain weight coefficients k i,j , where i,j are the serial numbers of the vertices from
Vm , which are connected by the edge Ei,j m in the graph G , m∈1..z, i,j∈1..n. Units of weight
m
 3. Problem Formulation
A multilayer data transmission network (MDTN) or a network model is a collection
of several layers representing a pool of undirected graphs 𝐺𝑚 , 𝑚1. . 𝑧 , where z is the
Computation 2023, 11, 74 number of graphs in the model, built on a set of a vertices Vm{v1,…,vn}. Edges of3graph of 14
𝑚 𝑚
Gm 𝐸𝑖,𝑗 have certain weight coefficients 𝑘𝑖,𝑗 , where i,j are the serial numbers of the vertices
𝑚
from Vm, which are connected by the edge 𝐸𝑖,𝑗 in the graph Gm, m1..z, i,j1..n. Units of
coefficients may varymay
weight coefficients depending on the subject
vary depending on thearea of the
subject area task. Thetask.
of the designations of the
The designations
graph
of thestructure have a composite
graph structure have a composite whichinnwhich
index, inindex, is the ordinal
n is the number of the vertex
ordinal number of the
and m is the number of the layer to which the given node belongs. For example,
vertex and m is the number of the layer to which the given node belongs. For example, the the third
vertex
third located in the second
vertex located layer will
in the second layerbewill
denoted as V 32as[11].
be denoted Graphical
V32 [11]. visualization
Graphical of
visualization
the MDTN model is presented in Figure 1. The units of the parameter may
of the MDTN model is presented in Figure 1. The units of the parameter may vary de- vary depending
on the subject
pending area
on the of thearea
subject task.of the task.

Figure1.1.Original
Figure Originalmultilayer
multilayerdata
datatransmission
transmissionnetwork.
network.

TheThetask
taskisistotofind
findthe
theshortest
shortestpath
pathfor forthe
theresulting
resultinggraph
graphafter
afterthe
theoperation
operationofof
combining
combiningallallitsitslayers.
layers.
TheTheunion
unionofoftwo graphsG1Gand
twographs GG
1 and 2 2isiscalled graphGG
calledaagraph 3 ,3,which
whichconsists
consistsofofthe
theunion
union
ofofallallvertices
verticesand
andedges
edgesofofthe
theoriginal
originallayers
layers(1).(1).

G1 (V1𝐺,1E(𝑉 , 𝐸 ) ∪ 𝐺 (𝑉 , 𝐸 ) = 𝐺 (𝑉 , 𝐸3 )(=
1 )1 ∪ 1G2 (V22, E22) =2 G3 (V33 , E33 ) = V3(𝑉 = 𝑉∪
=3 V ∪ 𝑉 , 𝐸 = 𝐸 ∪ 𝐸 ),
1 1 V2 ,2E3 3= E11∪ E22),
(1)
(1)
Themerged
The mergedMDTN
MDTNshown shownininFigure
Figure1 1isispresented
presentedininFigure
Figure2.2.
In Figure 2, the edges are highlighted in different colors:
In Figure 2, the edges are highlighted in different colors: red for red forthe
thefirst
firstlayer,
layer,green
green
forthe
for thesecond,
second,and
andyellow
yellowforforthe
thethird
thirdlayer
layerofofthe
theconsidered
consideredMDTN,
MDTN,respectively.
respectively.AfterAfter
merging, the MDTN needs to be preprocessed. Let us consider
merging, the MDTN needs to be preprocessed. Let us consider the problem using the the problem using the ex-
ample of N =
example of N = 3. 3.
A path in an undirected graph layered structure is a sequence of interconnected
vertices P = {vl,m , . . . , vs,m }∈V, m∈{1, . . . ,z}, l,s∈{1, . . . ,n}. Such a path P is called a path
from v = vl,m to v0 = vs,m .
The weight coefficient of edge ei,j m is k m . The weight function (2), which maps the edges
i,j
to their corresponding weights, representing the set of real numbers, is known.

f : E → R. (2)
 Computation 2023, 11, 74 4 of 14

Then, the shortest path from vertex v to vertex v’ is the path Pmin = {vl,m , . . . , vs,m }∈V,
where vl,m = v, vs,m = v0, which satisfies the value objective function represented by
Formula (3).
s z
Computation 2023, 11, x FOR PEER REVIEW ∑ ∑ km
i,j → min . 4 of 14(3)
i = l m =1
i,j∈ P

Figure
Figure 2. 2. Merged
Merged multilayer
multilayer data
data transmission
transmission network.
network.

If all edges in the graph have unit weight, then the problem is reduced to determining
A path in an undirected graph layered structure is a sequence of interconnected ver-
the least number of traversed edges.
tices P = {𝑣𝑙,𝑚 , … , 𝑣𝑠,𝑚 }∈V, m{1,…,z}, l,s{1,…,n}. Such a path P is called a path from 𝑣 =
Multilayer data transmission networks can be widely used in supply chain manage-
𝑣𝑙,𝑚 to 𝑣 ′ = 𝑣𝑠,𝑚 .
ment in tasks of the logistics sphere. 𝑚The supply
𝑚 chain management process is a simulation
The weight coefficient of edge 𝑒𝑖,𝑗 is 𝑘𝑖,𝑗 . The weight function (2), which maps the
of the transportation of manufactured goods through intermediate warehouses to the final
edges to theirThe
consumer. corresponding
most commonweights, representing
and optimal theof
scheme setwork
of real numbers,
is the is time”
“just in known.system,
which is characterized by the following 𝑓:
features:
𝐸 → 𝑅. (2)
- Stable production output at each time step;
- Then, the shortest
Frequent pathoffrom
deliveries vertex
products invsmall
to vertex v’ is the path 𝑃𝑚𝑖𝑛 = {𝑣𝑙,𝑚 , … , 𝑣𝑠,𝑚 }∈V,
batches;
where
- 𝑣 = 𝑣, 𝑣
𝑙,𝑚 of supply
Lack = 𝑣′, which satisfies
𝑠,𝑚 with excess or deficiency.the value objective function represented by for-
mula (3).
For the transportation of products, it is possible to use multimodal logistics networks
𝑠 𝑧
using various types of transport. Thus, each MDTN layer can represent different types of
𝑚
transport, and the nodes of the∑ ∑in𝑘this
layer 𝑖,𝑗 →case
𝑚𝑖𝑛are
. the reference points for transporting (3)
𝑖=𝑙 𝑚=1
products with the possibility of𝑖,𝑗∊𝑃 reloading products from one type of transport to another,
and the edges are the links between them.
If all edges
Thus, whenin the graph have
optimizing unit weight,
MDTN, then the
the process problem isthe
of choosing reduced to determining
best route for products
thebetween
least number of traversed edges.
given points according to the established parameters takes place.
Multilayer data transmission networks can be widely used in supply chain manage-
ment in tasks ofand
4. Materials the Methods
logistics sphere. The supply chain management process is a simulation
of 4.1.
the transportation
MDTN Preprocessing of manufactured goods through intermediate warehouses to the final
consumer. The most common
Finding the shortest and
path will optimal scheme
be carried of work
out using theisBellman–Ford
the “just in time” system,
algorithm. The
which is characterized by the following features:
modification of the algorithm consists in adding to it the preprocessing of the initial
- data,
Stable production
which output
will increase theatefficiency
each timeofstep;
finding optimal paths in multimodal transport
- networks.
FrequentThedeliveries
purposeofofproducts in small batches;
data preprocessing is to make sure that the graphs, which are
- constructed
Lack of supply with
from the excess or deficiency.
reachability matrix, are connected and do not have cycles and loops.
A graphical representation of the
For the transportation of products, designed system
it is possible tois shown
use using alogistics
multimodal data flow diagram
networks
(DFD) in Figure 3.
using various types of transport. Thus, each MDTN layer can represent different types of
transport, and the nodes of the layer in this case are the reference points for transporting
products with the possibility of reloading products from one type of transport to another,
and the edges are the links between them.
Thus, when optimizing MDTN, the process of choosing the best route for products
between given points according to the established parameters takes place.
 which will increase the efficiency of finding optimal paths in multimodal transpo
works. The purpose of data preprocessing is to make sure that the graphs, which a
structed from the reachability matrix, are connected and do not have cycles and lo
Computation 2023, 11, 74 graphical representation of the designed system is shown using a data 5 of 14 flow d
(DFD) in Figure 3.

Figure 3. DFD diagram of the Bellman–Ford algorithm with modification.

Figure 3. DFD diagram of the Bellman–Ford algorithm with modification.
A graph is called connected if there is only one connected component, i.e., when any
A graph
two vertices is (x,
of G called
y) areconnected
connected by if at
there
leastis only
one path. one connected
Otherwise, component,
the graph is called i.e., wh
disconnected [16]. An example of a connected graph is shown in
two vertices of G (x, y) are connected by at least one path. Otherwise, the graph is Figure 2.
To determine the connectivity, the algorithm of bypassing the graph model in breadth
disconnected [16]. An example of a connected graph is shown in Figure 2.
(breadth-first search) [17] is used. In this method, connectivity is checked by viewing all
To directions
possible determine from the connectivity,
a given vertex of thethe algorithm
graph with the transitionof bypassing thelevel
to a higher graph mo
breadth (breadth-first
at each iteration. Thus, the search) [17] starting
algorithm, is used.from
In this
the method,
source vertex connectivity
s, includes isin checked
the by
ing all all
search possible directions
nodes connected byfrom
edgesatogiven vertex
the given of the
vertex. Next, graph with theprocess
the transition transition
to to a
the obtained vertices is carried out and the browsing continues
level at each iteration. Thus, the algorithm, starting from the source vertex s, incluuntil the next attempt to
browse the next vertices does not find any new nodes of the graph.
the search all nodes connected by edges to the given vertex. Next, the transition p
Next, the check of the presence of cycles in the network structure is needed. If there
to
arethe obtained
such, vertices
it is possible is carried
that there are many out and the
shortest pathsbrowsing
from a given continues
vertex ofuntil the next a
the cycle
to
to browse
another, thesincenext
eachvertices
iteration does
of thenot
loopfind any the
reduces new nodes
length of the
of the pathgraph.
[18]. Brent’s
algorithm
Next,was theused
check to eliminate cycles. The
of the presence ofalgorithm
cycles inisthe chosen network due to structure
the fact that,
is on
needed.
average, it works about 36% faster than the Floyd algorithm and about 24% faster than the
are such, it is possible that there are many shortest paths from a given vertex of th
Pollard algorithm in solving a similar problem [19].
to another,
To describesince
theeach
Brent iteration
algorithm, of the loop
denote reduces
by M the the length
set consisting of the path
of m elements. Addi- [18]. Bre
gorithm was mapping
tionally, define used to f:M eliminate
→M and cycles. Theelement
an arbitrary algorithm is chosen
a0 belonging due
to the to the fact t
set under
average, it works about 36% faster than the Floyd
consideration. Next, consider the sequence of elements a 0 1 algorithm
, a , . . . , a n defined and about(4).
by equality 24% fast
the Pollard algorithm in solving an+i =af (similar
an ), n = problem
0, 1, . . . . [19]. (4)
To describe the Brent algorithm, denote by M the set consisting of
Since the considered set M is finite, the sequence (4) can be looped from some arbitrary
m elements
tionally,
moment. define mapping
In this case, f:M→M
the cycle and anfrom
can be started arbitrary element
any element aγ soa0that
belonging to the set
equality (5)
consideration.
is satisfied. Next, consider the sequence of elements a0, a1, …, an defined by equa
an = an+τ , for all n ≥ γ ≥ τ and τ ≥ 1, (5)
 𝑛+𝑖 𝑛 (4)

Since the considered set M is finite, the sequence (4) can be looped from some arbi-
trary moment. In this case, the cycle can be started from any element 𝒂𝜸 so that equality
(5) is satisfied.
Computation 2023, 11, 74 𝑎𝑛 = 𝑎𝑛+𝜏 , for all 𝑛 ≥ 𝛾 ≥ 𝜏 and 𝜏 ≥ 1, 6 of 14
(5)

where γ is the length of the approach to the cycle, τ is the length of the cycle [20].
whereThus,
γ isthe
theproblem
length ofof finding
the approachcycles in acycle,
to the graphτ structure is carried
is the length out with
of the cycle [20]. known
mappings f and start element a 0. The block diagram of the Brent method is shown in Figure
Thus, the problem of finding cycles in a graph structure is carried out with known map-
4.pings f and start element a . The block diagram of the Brent method is shown in Figure 4.
0

Figure 4. Block diagram of the Brent method.

Figure 4. Block diagram of the Brent method.
The main advantages of this algorithm are that no additional steps are required to
The main advantages of this algorithm are that no additional steps are required to
find and determine the correct cycle length τ, and the calculation function is called at each
find and determine the correct cycle length τ, and the calculation function is called at each
iteration step only once.
iteration step only once.
The performance speed of the method is O(g + τ), where g is the smallest index of the
sequence that is the beginning of the cycle, and τ is the length of the cycle. The MDTN
considered in Figure 1 after the removal of cycles in it is shown in Figure 5.
The Bellman–Ford algorithm is applied to the resulting graph to find the optimal path.
Computation 2023, 11, x FOR PEER REVIEW 7 of 14

The performance speed of the method is O(g + τ), where g is the smallest index of the
Computation 2023, 11, 74 7 of 14
sequence that is the beginning of the cycle, and τ is the length of the cycle. The MDTN
considered in Figure 1 after the removal of cycles in it is shown in Figure 5.

Figure 5. The original MDTN after removing cycles from it.

Figure 5. The original MDTN after removing cycles from it.
4.2. Finding the Optimal Route
The Bellman–Ford algorithm is applied to the resulting graph to find the optimal
The Bellman–Ford algorithm was created by American scientists as a distance vector
path.
routing (RIP) method that deals with transit sections between two network nodes through
which network packets are transmitted. Later, the algorithm was further developed for the
4.2. Finding the Optimal Route
problems of finding the minimum route of weighted network structures. Let us consider
stepThe Bellman–Ford
by step structure ofalgorithm was created by American scientists as a distance vector
the algorithm.
routing (RIP) method that deals
Step 1. In a given graph G with with transit
edge sections
weightsbetween two network
kij , the initial source nodes
vertexthrough
vi , from
which network packets are transmitted. Later,
which the shortest route will be searched, is determined. the algorithm was further developed for
the problems
Step 2. Anof array
finding of the minimum
distances d [0, route
1, . . . of, (qweighted
− 1)], where network structures.
q is the number of Let us con-
vertices in
sider step by step structure of the algorithm.
the graph, is specified. After the algorithm is executed, this array will contain the solution
Step
to the 1. In aof
problem given graph
finding theGoptimal
with edge
route.weights 𝒌𝒊𝒋the
Initially, , thearray
initial
is source vertex
initialized 𝒗𝒊 , from
according to
which
Formulathe(6).
shortest route will be searched, is determined.
Step 2. An array of distances d [0, 1, …, (q−1)], where q is the number of vertices in
the graph, is specified. = 0, where
d(v1 ) After v1 − startisvertex
the algorithm executed,of shortest
this arraypath search
will contain the solution (6)
d ( v ) = ∞.
to the problem of finding the optimal route. Initially, the array is initialized according to
i
formula 6.
Step 3. Computational core. The main operation of the algorithm is the operation of
𝑑(𝑣1 ) = 0, 𝑤ℎ𝑒𝑟𝑒 𝑣1 − start vertex of shortest path search
edge relaxation. Edge relaxation is the process of improving the value of the optimal route
(6)
obtained at the previous iteration of the 𝑑(𝑣 algorithm. At each iteration, the algorithm looks
𝑖 ) = ∞.
through all the edges of the graph and tries to relax along each edge (u, v) that has cost c.
Step 3. Computational
Algorithmically, core. Thefunction
the core relaxation main operation of the
is described in algorithm
formula 7. is the operation of
edge relaxation. Edge relaxation is the process of improving the value of the optimal route
obtained at the previous if iteration
d[v] > (of
d[uthe
] +algorithm.
c), else d[vAt ] =each
(d[u]iteration,
+ c ). the algorithm looks (7)
through all the edges of the graph and tries to relax along each edge (u, v) that has cost c.
For unreachable
Algorithmically, vertices
the core from function
relaxation the current node, thein
is described formulad(7.vi ) remains equal to
distance
infinity. A graphical representation of the relaxation process is shown in Figure 6.
Step 4. The algorithm if 𝑑[𝑣] > (𝑑[𝑢]
iteratively + 𝑐),the
refines elsevalue
𝑑[𝑣] of= (𝑑[𝑢] + 𝑐). d(v) at each step until
the function (7)
the process of relaxation of at least one edge occurs at the iteration.
For unreachable vertices from the current node, the distance 𝑑(𝑣𝑖 ) remains equal to
In its work, the Bellman–Ford algorithm can maximally perform (|V| − 1) iteration
infinity. A graphical representation of the relaxation process is shown in Figure 6.
over the relaxation of E edges. Therefore, its performance speed is defined as O(|V||E|).
Step 4. The algorithm iteratively refines the value of the function d(v) at each step
This parameter can be reduced by applying graph preprocessing by removing all cycles.
until the process of relaxation of at least one edge occurs at the iteration.
Computation 2023, 11, x FOR PEER REVIEW 8 of 14
In its work, the Bellman–Ford algorithm can maximally perform (|V|−1) ite
Computation 2023, 11, 74 over the relaxation of E edges. Therefore, its performance speed is defined as O(|V
8 of 14
In its work, the Bellman–Ford algorithm can maximally perform (|V|−1) iteration
This parameter can be reduced by applying graph preprocessing by removing all c
over the relaxation of E edges. Therefore, its performance speed is defined as O(|V||E|).
This parameter can be reduced by applying graph preprocessing by removing all cycles.

Figure 6. Graphical representation of the relaxation operation.

Figure 6. Graphical representation of the relaxation operation.
Figure 6. Graphical representation of the relaxation operation.
Figure 7 shows the optimal route between vertices V2V9 for the initial MDTN.
Figure 7 shows the optimal route between vertices V2 V9 for the initial MDTN.
Figure 7 shows the optimal route between vertices V2V9 for the initial MDTN.

Figure 7. Optimal route in a multilayer data transmission network.

4.3. Calculation
Figure 7. Optimalof the Coefficient
route of Admissibility
in a multilayer of Decisions
data transmission network.
The resulting solution is optimal. The optimal route is the only path by which you can
4.3.
get Calculation
from a given of the Coefficient
vertex of Admissibility
to the final of Decisions
one with minimal losses. However, there are tasks for
whichTheit resulting
is necessary to have
solution backup shortest
is optimal. routes.
The optimal Such
route is options
the onlywithpatha bycertain
whichdegree
you
of fault
can tendato
get from the weight
given vertex toofthe
thefinal
optimal
one route, therefore,
with minimal it is However,
losses. necessary there
to consider the
are tasks
Figure 7. Optimal route in a multilayer data transmission network.
possibility
for which it of forming a to
is necessary pool
haveof backup
paths that are not
shortest optimal,
routes. Such but close in
options costa certain
with to them.degree
Let us
call
of such
fault routes
tend admissible.
to the weight of the optimal route, therefore, it is necessary to consider the
4.3. Calculation
The formation
possibility of the
of forming a Coefficient
of admissible
pool ofthat
of pathsroutesAdmissibility
occurs
are afterofthe
not optimal, Decisions
determination
but close in cost toofthem.
the optimal
Let us
path and
call such implies
The routes
resulting a deviation
admissible. from the cost by a given value ε, which is
solution is optimal. The optimal route is the only path by whic set experimentally
depending on the conditions
The formation of admissibleof the problem
routes occursbeing
aftersolved [11].
the determination of the optimal path
can get from
The general
a given
view of
vertex
the the
to the
calculation
final one with minimal losses. However,
Kdop is shown
there are
and implies a deviation from cost by aofgiven
the coefficient of admissibility
value ε, which is set experimentally de-
for
inwhich
Formula
pending
it (8).
on the
is necessary to have backup shortest[11].
conditions of the problem being solved
routes. Such options with a certain d


of fault
Thetend
general to view
the weight of the
of the calculation optimal
Kdop of=the Proute,
opt ,
ε%coefficient therefore,
of admissibilityit is 𝐾necessary
𝑑𝑜𝑝 is shown
to consid
(8)
possibility
in Formula (8).of forming a pool of paths that are not optimal, but close in cost to them.
4.4. Algorithm Operation with Preferences
call such routes admissible. 𝐾𝑑𝑜𝑝 = 𝜀%(𝑃𝑜𝑝𝑡 ), (8)
In practice, there are cases in which the obtained optimal solution is P , where
min
The formation of admissible routes occurs after the determination of the optima
there is a repeated change in the ownership of layers in a multilayer data transmission
and implies
network. In a deviation
the tasks of from themultimodal
managing cost by alogistics
given value
systems,ε, this
which is set
means thatexperimental
when
4.4. Algorithm Operation with Preferences
pending on the
transporting conditions
products from the of producer
the problem to thebeing solved
consumer (from[11].
the start vertex to the
In practice, there are cases in which the obtained optimal solution is Pmin, where there
finalThe
one),general
there is aview
repeated
of thechange in types ofof
calculation transport.
the In this case,
coefficient when choosing the
of admissibility 𝐾𝑑𝑜𝑝 is s
isoptimal
a repeated change
(efficient) in theit ownership
route, is advisableoftolayers in a account
take into multilayer
thedata
cost transmission
of the double network.
operation
in Formula (8).
“unloading-loading”, which includes not only material and time resources but also the
𝐾𝑑𝑜𝑝 = 𝜀%(𝑃𝑜𝑝𝑡 ),

4.4. Algorithm Operation with Preferences

 ing-loading”, which includes not only material and time resources but also the costs asso-
ciated with the peculiarity of the goods being transported, such as transportation temper-
ature, shelf life, etc. In the case, when the additional costs exceed the costs associated with
transportation when changing the layer (type of transport), a decision is made to remain
Computation 2023, 11, 74 in the priority layer for product transportation. Moreover, to maintain the efficiency of 9 of 14
such a network, it is necessary that the selected path is in the pool of acceptable routes,
with a set reachability coefficient for the problem being solved.
For the
costs considered
associated with example,
the peculiarity the optimal route being
of the goods was obtained
transported, as a such
resultasoftransportation
applying
the temperature,
Bellman–Ford shelf life, etc. In the case, when the additional costs exceed the costsofassociated
algorithm using data preprocessing (Figure 6) and consists a se-
quence of nodes P = {v 22, v21, v24, v35, v17, v19}. Moreover, according to the solution of the
with transportation when changing the layer (type of transport), a decision is made to
problem,
remaintransportation
in the prioritybetween layer forvproduct
22v21 and v21v24 nodes is carried out along the second (m
transportation. Moreover, to maintain the efficiency
= 2)ofMDTN layer, and therefore
such a network, it is necessary that the belongs to the second
selected type
path of the
is in transport;
pool of transportation
acceptable routes,
between
with anodes v24v35 passes
set reachability through the
coefficient third
for the (m = 3) being
problem layer, solved.
and between nodes v35v17 and
v17v19 through
For thethe first (m =example,
considered 1) layer,the respectively.
optimal route In such a route, all
was obtained as athree
resultmodes of
of applying
transport (layers) are used to transport products in the supply
the Bellman–Ford algorithm using data preprocessing (Figure 6) and consists of a sequence chain, and the third type of
transport is used only once: for transportation between v v nodes.
of nodes P = {v22 , v21 , v24 , v35 , v17 , v19 }. Moreover, according to the solution of the problem,
34 35 In this case, it is nec-
essary to check the between
transportation feasibilityv22 ofvusing the third transportation layer in the resulting prod-
21 and v21 v24 nodes is carried out along the second (m = 2)
uct MDTN
route. The layer,verification
and therefore algorithm
belongs will consist
to the secondof the
typefollowing
of transport; steps:transportation between
Step 1.
nodes v24Checking
v35 passesfor a connection
through the third between
(m = 3)neighboring
layer, and between vertices nodes
at the vlevel
35 v17of other
and v17 v19
layers. If there
through theare firstno(msuch = 1)connections,
layer, respectively. then the Inoriginal layer all
such a route, of the
three problem
modes from the
of transport
optimal route
(layers) areremains
used to involved.
transport products in the supply chain, and the third type of transport
is used only is
Step 2. If it possible
once: to use transportation
for transportation between onv34alternative
v35 nodes.layers In thisbetween
case, it isthenecessary
consid- to
eredcheck
nodes, thethen the layers
feasibility used in
of using thethe nodes
third adjacent to them
transportation layerare checked.
in the resulting From the pool
product route.
of alternative
The verification layers,algorithm
the layer will withconsist
the lowestof the transportation
following steps: cost is selected.
So, Step
for example,
1. Checking for the foroptimal
a connection solution shownneighboring
between in Figure 6, vertices
there is an alternative
at the level of to other
layers.from
transport If there
the vare no such
24 vertex connections,
to the v35 vertex then both the
alongoriginal
the first layer
andof the problem
second layers. How-from the
optimal
ever, preferenceroutewill remains
be given involved.
to the first layer, because the cost of transportation through
Step 2. If it is
it will be lower than through the possible toalternative
use transportation
second. The on alternative
new product layers between the
transportation consid-
route
withered nodes,
a given then thecoefficient
tolerance layers used in the
of 10% nodes
after adjacent
applying thetopreference
them are checked.
algorithmFrom the pool
is shown
of alternative
in Figure 8. layers, the layer with the lowest transportation cost is selected.
StepSo, 3. for
Anexample,
additional forverification
the optimalof solution shown in Figure
the membership of the6,found
there solution
is an alternative
of the to
transport from the v vertex to the v vertex both along
problem is carried out, taking into account the set admissibility coefficient. If this param-
24 35 the first and second layers. How-
eter is satisfied, a recommendation is issued to use not the optimal route, but the path it
ever, preference will be given to the first layer, because the cost of transportation through
will beaslower
obtained a result than of through
the algorithm the alternative second. The
with preferences. If thenew product
found transportation
alternative route does route
with a given tolerance coefficient of 10% after applying
not satisfy the given tolerance coefficient, then the transportation route remains un- the preference algorithm is shown
in Figure 8.
changed.

Figure
Figure 8. The
8. The route
route of transportation
of transportation of of products
products afterthe
after theoperation
operationof
ofthe
the algorithm
algorithm with
with preferences
prefer-
with
ences a given
with tolerance
a given tolerancecoefficient of of
coefficient 10%.
10%.

Step 3. An additional verification of the membership of the found solution of the

problem is carried out, taking into account the set admissibility coefficient. If this parameter
is satisfied, a recommendation is issued to use not the optimal route, but the path obtained
as a result of the algorithm with preferences. If the found alternative route does not satisfy
the given tolerance coefficient, then the transportation route remains unchanged.
Thus, the use of an algorithm with preferences makes it possible to modify the found
optimal solution by removing single transportations between MDTN layers from it, which
is achieved by replacing these layers with alternative ones while maintaining that the
solution of the problem belongs to the pool of feasible solutions.
 Thus, the use of an algorithm with preferences makes it possible to modify the found
optimal solution by removing single transportations between MDTN layers from it, which
is achieved by replacing these layers with alternative ones while maintaining that the so-
lution of the problem belongs to the pool of feasible solutions.
Computation 2023, 11, 74 10 of 14
5. Results
To implement and conduct a comparative analysis of the considered methods, an HP
5.Laptop
Results15s-fq2002ur personal computer was used. The computer is equipped with a
quad-core Intel Core
To implement andi5-1135G7
conductprocessor
a comparativeand the Microsoft
analysis Windows
of the 10 version
considered 2004 an
methods, op-
erating
HP Laptopsystem. As the programming
15s-fq2002ur personal computer language waswas chosen
used. Thethe object-oriented
computer C#. with
is equipped
a quad-core Intel Core
The software i5-1135G7
allows to enterprocessor
the data set and asthe Microsoft
a text Windows
file or generate 10 version
data 2004
randomly, so
operating
the data setsystem. As the programming
is synthetic. For the study, language wasconsisting
the data set, chosen theofobject-oriented C#. and a
five graph layers
The software
variable allows towith
graph dimension enter the data setofasfifteen
a maximum a textvertices,
file or generate
was used. data randomly,
When so
construct-
the data set is synthetic. For the study, the data set, consisting of five
ing graphs, an upper estimate of the time complexity was chosen. In practice, the problem graph layers and a
variable graph dimension with a maximum of fifteen vertices, was used. When
can be applied to a public data set, where certain settlements or logistics centers can act as constructing
graphs,
vertices.an upper estimate of the time complexity was chosen. In practice, the problem
can beTo applied
test thetodeveloped
a public data set, where
system, certain settlements
it is necessary to apply theor logistics centers
reachability can act
coefficient 𝐾𝑑 ,
as vertices.
which characterizes the degree of connectivity of graphs and determines the proportion
To test
of filling thethe developedmatrix
reachability system, withit isnon-zero
necessary to apply[21].
elements the reachability coefficient Kd ,
which characterizes the degree of connectivity of graphs and determines the proportion of
𝑘𝑛𝑢𝑙𝑙
filling the reachability matrix with non-zero 𝐾𝑑 =elements ,[21].
𝑁2 − 𝑁 (9)
k 2
Kd = 2null , (9)
where N is the dimension of the reachability N −matrix,
N knull is the number of non-zero ele-
2
𝑁2 −𝑁
ments of this matrix, is the total number of elements in the matrix.
where N is the dimension2of the reachability matrix, knull is the number of non-zero elements
The dependence
N2 − N
of the performance of the Bellman–Ford algorithm on the values of
ofthe
this matrix,
MDTN graph is the total number
2 reachability of elements
coefficient in the
was studied, thematrix.
graph was obtained, and it is
The dependence
shown in Figure 9. of the performance of the Bellman–Ford algorithm on the values of
the MDTN graph reachability coefficient was studied, the graph was obtained, and it is
Testing was carried out on a multilayer data transmission network consisting of five
shown in Figure 9.
layers, each of which had no more than ten nodes.

0.7

0.6

0.5
Time, sec

0.4

0.3

0.2

0.1

0
0.2 0.5 0.75
Reachability coefficient

Bellman-Ford algorithm Bellman-Ford algorithm with modification

Figure9.9.Dependence
Figure Dependenceofofthe
theperformance
performanceofofalgorithms
algorithmson
onthe
thereachability
reachabilitywithout
withoutconstruction
constructionofof
feasible routes.
feasible routes.

Testing was carried out on a multilayer data transmission network consisting of five
layers, each of which had no more than ten nodes.
Figures 10 and 11 show the dependence of the performance of the algorithms on
the reachability coefficient of the graph structure, taking into account the construction of
feasible routes with an average admissibility coefficient of 10% and 30%, respectively.
 Computation 2023, 11, x FOR PEER REVIEW 11 of 14

Computation 2023, 11, x FOR PEER REVIEW 11 of 14

Figures 10 and 11 show the dependence of the performance of the algorithms on the
Computation 2023, 11, 74 11 of 14
reachability
Figures coefficient of thethe
10 and 11 show graph structure,oftaking
dependence into account
the performance of the
the construction
algorithms onofthe
fea-
sible routes with an average admissibility coefficient of 10% and 30%, respectively.
reachability coefficient of the graph structure, taking into account the construction of fea-
sible routes with an average admissibility coefficient of 10% and 30%, respectively.
1
0.9
1
0.8
0.9
0.7
0.8

sec sec
0.7
0.6

Time,
0.6
0.5
Time,
0.5
0.4
0.4
0.3
0.3
0.2
0.2
0.1
0.10
0 0.2 0.5 0.75
0.2 0.5 0.75
Reachability coefficient
Reachability coefficient
Bellman-Ford algorithm Bellman-Ford algorithm with modification
Bellman-Ford algorithm Bellman-Ford algorithm with modification
Figure10.
Figure 10.Dependence
Dependenceof ofthe
theperformance
performanceof
ofalgorithms
algorithmson
onthe
thereachability
reachabilityof
ofthe
theMDTN
MDTNwith
withaa
Figureadmissibility
route 10. Dependence of the performance
coefficient of 10%. of algorithms on the reachability of the MDTN with a
route admissibility coefficient of 10%.
route admissibility coefficient of 10%.

3.5
3.5
3
3
2.5
2.5
secsec

2
2
Time,
Time,

1.5
1.5
1
1
0.5
0.5
0
0
0.2 0.5 0.75
0.2 0.5 0.75
Reachability coefficient
Reachability coefficient

Bellman-Ford algorithm Bellman-Ford algorithm with modification

Bellman-Ford algorithm Bellman-Ford algorithm with modification

Figure 11. Dependence of the speed of algorithms on the coefficient of reachability of the MDTN
Figure 11.Dependence
Figure Dependence of of the
the speed
speed of
of algorithms
algorithms on
on the
the coefficient
coefficient of
of reachability
reachabilityof
ofthe
theMDTN
MDTN
with a11.
route admissibility coefficient of 30%.
with a route admissibility coefficient of 30%.
with a route admissibility coefficient of 30%.
The dependence of
Thedependence
dependence of consumed
consumed memory
memory volumeson on a numberofofMDTN MDTN layerswas was
The of consumed memory volumes
volumes on aa number
number of MDTNlayers layers was
studied. Based
studied.Based
Basedononon the comparative analysis, it can be concluded that the amount ofof
studied. thethe comparative
comparative analysis,
analysis, it canit be
can be concluded
concluded that thethat the amount
amount of memory
memory consumed
memory consumed by the
by theand original
original and modified
and modified algorithms
algorithms differs slightly.
differs slightly.
consumed by the original modified algorithms differs slightly.
Figure
Figure12 12 shows
12shows a comparative
shows aa comparative analysis
comparative analysis of the operation
analysis of the operation of themodified
modified algorithm
Figure operation of of the
the modified algorithm
algorithm
for finding
forfinding
findingthethe shortest
theshortest route
shortest route
route inin MDTN
in MDTN without post-processing
MDTN without post-processing the result and usingthe the
for post-processing the the result
resultand
andusing
using the
algorithm
algorithmwithwith preferences.
withpreferences.
preferences.Based Based
Basedon on
onthethe test
thetest results,
testresults,
results, it can be seen that the processing ofof
algorithm it it
cancan
bebe seen
seen that
that thethe processing
processing of the
the results
the results
results obtained
obtained
obtained requires
requires
requires additional
additional
additional time
timetime
costs, costs,
costs, on average,
on average,
on average, depending
depending
depending on
on the
on the the reacha-
reacha-
reachability
bility
bility coefficient,
coefficient, 1.5–2
1.5–2 times
times higher
higher than
than the
the algorithm
algorithm without
without
coefficient, 1.5–2 times higher than the algorithm without post-processing. post-processing.
post-processing.
Computation 2023,11,
Computation2023, 11,74x FOR PEER REVIEW 12
12 of
of1414

3.5
3
2.5

Time, sec
2
1.5
1
0.5
0
0.2 0.5 0.75
Reachability coefficient

Without the use of post-processing results

Using an Algorithm with Preferences

Figure 12. Comparative analysis of the modified version of the Bellman–Ford algorithm without
Figure 12. Comparative analysis of the modified version of the Bellman–Ford algorithm without
post-processing of results and with it.
post-processing of results and with it.

Theresult
The resultofofthe
thework
workwas
wasthe
theconstruction
constructionof ofvarious
variouscompeting
competingoptions
optionsforfortrans-
trans-
porting products from the start vertex to the final one in order to build promising
porting products from the start vertex to the final one in order to build promising routes for routes
forformation
the the formation
of notofonly
not an
only an optimal
optimal and efficient
and efficient solutionsolution
but alsobut also
close in close
cost tointhem.
cost to
them.
6. Conclusions
6. Conclusions
To model the process of a multimodal transport network, it is possible to use a
To model
multilayer the process ofnetwork
data transmission a multimodal transport
consisting network,
of several graphsit issuperimposed
possible to use onaeach
mul-
tilayer
other, data transmission
representing network
various types consisting
of product of several graphs superimposed on each
transportation.
other,
In representing various route
finding the optimal types in of the
product transportation.
multilayer data transmission network, the best
In finding
performance wasthe optimal
shown by theroute in the multilayer
Bellman–Ford datawithout
algorithm transmission network,because
preprocessing, the best
additional
performance methods require
was shown byadditional time costs.
the Bellman–Ford However,
algorithm when calculating
without preprocessing, admissible
because
routes, the original algorithm showed slow performance compared
additional methods require additional time costs. However, when calculating admissible to the modified version.
Thus,
routes,with
thean admissibility
original algorithm coefficient
showed of 10%,
slow the time gain
performance was 23%
compared toof
thethe time of the
modified ver-
original algorithm,
sion. Thus, with an and with an admissibility
admissibility coefficientcoefficient
of 10%, theof time
30%, gain
the modified
was 23%version worksof
of the time
two
the times
originalfaster than theand
algorithm, original
with an Bellman–Ford
admissibility algorithm
coefficient onofa 30%,
graphthe with a reachability
modified version
coefficient
works twooftimes 0.75.faster
This than
happened due toBellman–Ford
the original the fact that with an increase
algorithm in thewith
on a graph number of
a reach-
layers, the number of cycles in the structure also increases, without
ability coefficient of 0.75. This happened due to the fact that with an increase in the num- removing which the
Bellman–Ford
ber of layers, the algorithm
numberisofexecuted
cycles infor theeach admissible
structure route. without removing which
also increases,
Preprocessing of the MDTN let to uniquely
the Bellman–Ford algorithm is executed for each admissible identify theroute.
only route between given
nodes,Preprocessing
which makes it possible to speed up the process
of the MDTN let to uniquely identify the only of program execution. At the same
route between given
time,
nodes,forwhich
weakly connected
makes graph
it possible structures,
to speed up thethe reachability
process of programcoefficient
execution.of which
At thedoes
same
not
time, for weakly connected graph structures, the reachability coefficient of which does(the
exceed 0.3, the performance time of the algorithms does not differ significantly not
difference
exceed 0.3, is less than 3%), however,
the performance time of a direct relationship
the algorithms between
does the increase
not differ in the(the
significantly model
dif-
reachability
ference is less coefficient
than 3%), and the increase
however, in the
a direct time spentbetween
relationship for searching has been
the increase in revealed.
the model
The amount of memory consumed when using a modified algorithm
reachability coefficient and the increase in the time spent for searching has been revealed. is approximately
7% higher than when
The amount using theconsumed
of memory standard Bellman–Ford
when using a algorithm, however, in
modified algorithm is modern
approxi-
realities, this difference is insignificant.
mately 7% higher than when using the standard Bellman–Ford algorithm, however, in
The introduced
modern realities, thisroute admissibility
difference coefficient makes it possible to determine not only
is insignificant.
the optimal path but also those close to it with a given accuracy, which allows finding an
The introduced route admissibility coefficient makes it possible to determine not only
additional pool of possible solutions and increases the practical application of the problem
the optimal path but also those close to it with a given accuracy, which allows finding an
under consideration.
additional pool of possible solutions and increases the practical application of the problem
under consideration.
Computation 2023, 11, 74 13 of 14

The possibility of modifying the obtained optimal route for a multilayer data transmis-
sion network within a given admissibility coefficient is considered. Despite the required
additional time spent on the post-processing of the result, the algorithm with preferences
makes it possible to obtain the optimal solution not only from the side of the cost of
transportation but also to take into account the possibility and expediency of loading and
unloading operations of transported products at MDTN nodes.
The just-in-time concept involves minimizing stocks of goods in warehouses and
rationalizing the use of various types of transport. In the future, it is planned to introduce
additional indicators to model the operation of a multilayer data transmission network,
such as the coefficient of warehouse turnover and the percentage of occupancy of vehicles
at each stage of loading–unloading goods.

Author Contributions: O.T. and A.S. contributed to this work in the formulation of the mathematical
model, and the process of experimentation, and they analyzed and interpreted the results. M.S. and
T.B. participated in the editing and writing of the document; they conducted a literature review based
on the importance of the put-away routing problem, and in addition, they analyzed the state-of-
the-art models applied to the solutions of these problems. All authors have read and agreed to the
published version of the manuscript.
Funding: This research received no external funding.
Data Availability Statement: All data generated or analyzed during this study are included in this
published article.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Oliveira, J.B.; Jin, M.; Lima, R.S.; Kobza, J.E.; Montevechi, J.A.B. The role of simulation and optimization methods in supply chain
risk management: Performance and review standpoints. Simul. Model. Pract. Theory 2019, 92, 17–44. [CrossRef]
2. Chueanun, S.; Suwandechochai, R. Effect of Additional Order in Two-Stage Supply Chain Contract under the Demand Uncertainty.
Computation 2021, 9, 37. [CrossRef]
3. Gomez-Montoya, R.A.; Cano, J.A.; Cortes, P.; Salazar, F. A Discrete Particle Swarm Optimization to Solve the Put-Away Routing
Problem in Distribution Centres. Computation 2020, 8, 99. [CrossRef]
4. Liu, D.; Hu, X.; Jiang, Q. Design and optimization of logistics distribution route based on improved ant colony algorithm. Optik
2023, 273, 170405. [CrossRef]
5. Montoya, O.D.; Molina-Cabrera, A.; Grisales-Noreña, L.F.; Hincapie, R.A.; Granada, M. Improved Genetic Algorithm for
Phase-Balancing in Three-Phase Distribution Networks: A Master-Slave Optimization Approach. Computation 2021, 9, 67.
[CrossRef]
6. Martynyuk, M.V.; Sokolova, E.S.; Dmitriev, D.V.; Tyurin, A.I. Matrix Method of Calculation for Simulation of Distribution Electric
Networks of Medium Voltage. Int. J. Appl. Eng. Res. (IJAER) 2017, 12, 15066–15072.
7. Confessore, G.; Galiano, G.; Liotta, G.; Stecca, G. A Production and Logistics Network Model with Multimodal and Sustainability
Considerations. Proce. CIRP 2013, 12, 342–347. [CrossRef]
8. Feng, X.; Song, R.; Yin, W.; Yin, X.; Zhang, R. Multimodal transportation network with cargo containerization technology:
Advantages and challenges. Transp. Policy 2023, 132, 128–143. [CrossRef]
9. Elbert, R.; Müller, J.P.; Rentschler, J. Tactical network planning and design in multimodal transportation–A systematic literature
review. Res. Transp. Bus. Manag. 2020, 35, 100462. [CrossRef]
10. Sedeño-noda, A.; Colebrook, M. A biobjective Dijkstra algorithm. Eur. J. Oper. Res. 2019, 276, 106–118. [CrossRef]
11. Nasiboglu, R. Dijkstra solution algorithm considering fuzzy accessibility degree for patch optimization problem. Appl. Soft Comput.
2022, 130, 109674. [CrossRef]
12. Gdansky, N.I.; Kulikova, N.L.; Chumakova, E.V. Exact solution of the problem of finding the minimum acyclic path in weighted
graphs containing edges of negative weight. Softw. Prod. Syst. 2018, 31, 260–267.
13. Harabor, D.; Grastien, A. The JPS pathfinding system. In Proceedings of the Fifth Annual Symposium on Combinatorial Search,
Niagara Falls, ON, Canada, 19 July 2012; pp. 207–208.
14. Harabor, D.; Grastien, A. Improving Jump Point Search. In Proceedings of the Association for the Advancement of Artificial
Intelligence, Arlington, VA, USA, 11 July 2015; Australian National University College of Engineering and Computer Science:
Sydney, Australia, 2015.
15. Maksimova, E.I. Comparison of the quality of the results of the “A star” algorithm and its modifications for the road network
when choosing a route taking into account the direction of movement at the intersection. Bull. Sci. Sib. 2018, 4, 117–122.
16. Diestel, R. Graph Theory, 5th ed.; Springer: Berlin/Heidelberg, Germany, 2017; p. 429.
Computation 2023, 11, 74 14 of 14

17. Kirk, D.; Hwu, W. Programming Massively Parallel Processors; Morgan Kaufmann is an Imprint of Elsevier: Waltham, MA, USA,
2016; p. 576.
18. Erzin, A.I.; Kochetkov, Y.A. Routing Tasks: Textbook. Manual; Novosibirsk State University: Novosibirsk, Russia, 2014; p. 95.
19. Brent, R.P. An Improved Monte Carlo Factorization Algorithm. Nord. Tidskr. Inf. Sbehandling (BIT) 1980, 20, 176–184. [CrossRef]
20. Nesterenko, A.Y. Algorithms for finding cycle lengths in sequences and their applications. Fundam. Appl. Math. 2010, 6, 109–122.
21. Stepanenko, M.A. Development of Models and Algorithms for Optimizing Diagnostic Procedures on Graph Models of Technical
Systems. Dissertation of the Candidate of Technical Sciences, Nizhny Novgorod State Technical University n.a. R.E. Alekseev,
Nizhny Novgorod, Russia, 2007.

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.