Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

ISSN 0005-1179, Automation and Remote Control, 2019, Vol. 80, No. 7, pp. 1358–1367. c Pleiades Publishing, Ltd.

, 2019.
c The Author(s), 2018, published in Matematicheskaya Teoriya Igr i Ee Prilozheniya, 2018, No. 1, pp. 99–115.
Russian Text 

MATHEMATICAL GAME THEORY AND APPLICATIONS

Coalition-Formation Problem
for Cooperative Inventory Routing Game
V. A. Shirokikh∗,a and E. A. Lezhnina∗,b

St. Petersburg State University, St. Petersburg, Russia
e-mail: a shva.erivel@gmail.com, b e.lezhnina@spbu.ru
Received November 11, 2017
Revised November 11, 2017
Accepted November 8, 2018

Abstract—This paper studies stability of carrier coalitions in a cooperative inventory routing


game (CIRG). Difficulty of this study is not only in a computational complexity of the class of
routing problems, but also in the task of constructing a characteristic function, since heuristic
solutions that are usually used in routing problems can’t guarantee the subadditivity property
in the general case. In its turn, violation of subadditivity can lead to instability of a coalition,
because a player could get more profit in a different coalition or individually. To solve routing
problems, Adaptive large neighborhood search (ALNS ) and its modification with the Dynamic
adaptation method, DALNS, are used in this work. A special Direct coalition induction al-
gorithm (DCIA) is used to construct a subadditive characteristic function, and four different
concepts of cooperative game solutions are considered. The analysis of extensive computational
experiments allows to illustrate the dependence of the stability of a grand coalition on such
factors as a routing algorithm, an algorithm for a characteristic function construction, and a
solution concept for a cooperative game.

Keywords: inventory routing problem (IRP ), cooperative inventory-routing game (CIRG), char-
acteristic function, heuristic algorithm
DOI: 10.1134/S0005117919070129

1. INTRODUCTION
Cooperation between carriers, who provide goods delivery service, might result in significant
benefits for participants if a mutual solution is obtained. Currently, there are a number of studies
on the cooperation of shippers. However, the overwhelming majority of them are devoted to explor-
ing opportunities for cooperation, which cover the decision-making process or the cost allocation
methods for the large coalition [11], but do not address routing problems. Several studies address
the problem of routing and create cooperative game models [16]. However, these studies do not
take into account the heuristic nature of solutions [14, 15] or work directly with optimal [13]. Such
an approach threatens possible collaboration due to the problem of solution instability when using
heuristic methods. This problem was first solved by Zakharov and Shchegryayev [18]. Routing-
related problems are complex combinatorial optimization problems, they belong to the NP-hard
class, and arise in large-scale systems which are modern distribution networks. An Inventory
Routing Problem (IRP ) is a joint problem of route construction and inventory management over
timeframe T consisting of periods t = 1, . . . , T . Since the present work focuses mainly on a cooper-
ative game aspects, therefore, for the sake of simplicity, we shall take for a basis of this work IRP
model, also considered, for example, by Archetti et al. [1]: one vehicle, one depot, homogeneous
product, deterministic demand, no time windows, etc. An extensive review of contemporary IRP
models can be found in papers by Coelho et al. [5] and Archetti et al. [2].

1358
COALITION-FORMATION PROBLEM 1359

The theory of cooperative games is widely used to simulate the interactions of several participants
in different situations. This theory provides various mathematical methods for the calculation of
individual utility for each player in the coalition, but before using them, it is necessary to construct a
characteristic function of the game. The main condition imposed on this function is subadditivity
(if the goal is to minimize the cost function), which actually might not be satisfied, when the
heuristic algorithm is used to obtain cooperative solutions. In such cases, situations are possible
when it is unprofitable for players to remain in a certain coalition: they can get more benefits by
working on their own or joining another coalition [12]. The stability of the coalition depends on
the chosen solution concept. The article explores the stability of the maximum coalition for the
four solution concepts.
2. MODEL
Suppose there is a complete graph G = ({M, N }, A), where M = {1, . . . , m} is the set of vertices
in which the suppliers are located; N = {1, . . . , n} is the set of vertices in which consumers are
located; A is the set of arcs connecting vertices.
Each supplier serves a specific group of consumers Ni ⊆ N and should solve a private IRP
problem on a reduced graph Gi = ({{i}, Ni }, A). The following conditions are assumed: if i1 , i2 ∈ M
and i1 = i2 , then Ni1 ∩ Ni2 = ∅.
Travel cost along arc (i, j) ∈ A is equal to cij > 0. Planning horizon consists of T periods. In
the beginning of each period amount of ri of product is delivered to the site by the supplier i, and
in the end, amount of di are delivered to each consumer j from the storage. Fleet of each carrier
includes one vehicle. During each period, a vehicle can perform one route that starts and ends in
the depot. Capacity of each vehicle is limited by Q.
Customers are able to store the product, thus having varying inventory level Ijt , t = 1, . . . , T .
Storage capacities are limited by values Uj . Holding costs per capacity unit is hj .
Routes and delivery amounts are variables in the IRP, and the goal is to minimize total trans-
portation and holding costs during time horizon.
Assume that carriers are able to form coalitions S ⊆ M , uniting their sets of customers. Thus,

each coalition serves many consumers NS = Ni .
i∈S
The following conditions must be satisfied:
NS1 ∩ NS2 = ∅, if S1 ∩ S2 = ∅.

Each coalition would have to solve special multi-depot case of IRP.


Suppose that xijt is a binary variable, equal to one if any route at period t passes the arc (i, j),
and zero otherwise; qij is integer variable, which represents amount of delivery to customer i at
period t. Next, new real variables sit are introduced, reflecting the total value of the delivered cargo
in the current route, including customer i in period t. An important function of these variables is
to ensure the absence of closed cycles that are not related to the point of departure (depot). The
initial stock level for consumers and carriers Ii0 = Ii0 = const for all i ∈ N . Thus, the IRP for the
coalition S can be defined in the form of linear program as follows:
T
   T
 
cij xijt + hi Iit → min, (2.1)
t=1 i∈S∪NS j∈S∪NS t=1 i∈S∪NS

Iit = Ii,t−1 + ri − sjt xjit , ∀i ∈ S ∪ NS , t = 1, . . . , T, (2.2)
j∈NS

Ijt = Ij,t−1 − dj + qjt , j ∈ NS , t = 1, . . . , T, (2.3)


0 ≤ Iit ≤ Ui , ∀i ∈ S ∪ NS , t = 1, . . . , T, (2.4)

AUTOMATION AND REMOTE CONTROL Vol. 80 No. 7 2019


1360 SHIROKIKH, LEZHNINA
 
xijt = xjit ≤ 1, ∀i ∈ S ∪ NS , t = 1, . . . , T, (2.5)
j∈NS j∈NS
 
xijt = xjit ≤ 1, i ∈ S, t = 1, . . . , T, (2.6)
i∈NS i∈NS
qit ≤ sit ≤ Q, ∀i ∈ NS , t = 1, . . . , T, (2.7)
sit − sjt + Qxijt ≤ Q − qjt, ∀i, j ∈ NS , i = j, t = 1, . . . , T. (2.8)

The target function (2.1) represents the sum of total transportation and total holding costs.
Constraints (2.2) and (2.3) represent the inventory dynamic at depot and at customers. Formu-
las (2.4) and (2.7) describe the limits on the size of the depot and transport for transportation.
The left parts in (2.5) and (2.6) assure routes continuity, while the right parts hold for single visits
of each customer (2.5) and single vehicles for each carrier (2.6). Additionally, the restriction (2.7)
ensures that the capacity of the vehicle is not violated. To define a cooperative game, we need to
define a set of players and a characteristic function for the game. Denote by M the set of players
in the case of CIRG.
Definition 2.1. Characteristic function is a function c : 2M → R, where 2M is the set of all pos-
sible coalitions of players satisfying the condition c(∅) = 0.
It is well known that the necessary condition for cooperation of players which try to minimize
their costs is the subadditivity of the characteristic function of cooperative game. Otherwise, players
will not want to join the coalition. Thus, it is necessary to construct a subadditive characteristic
function such that

∀S, T ⊂ M : S ∩ T = ∅, c(S ∪ T ) ≤ c(S) + c(T ). (2.9)

It is easy to notice that in CIRG for any two non-intersected coalitions S and T the exact
minimum value fopt of the target function (2.1) obviously satisfies the condition
     
fopt S ∪ T ≤ fopt S + fopt T . (2.10)

Unfortunately, in reality, one cannot get these exact minimal values of an target function in
large-scale problems because of the computational complexity of the IRP and heuristic algorithm
had to be used. For each coalition S, an approximate value f h (S) can be found using a heuristic
algorithm.
Such heuristic solution does not guaranteed to be the exact optimal one, i.e., the following
condition occurs:

fopt (S) ≤ f h (S). (2.11)

Combining (2.10) and (2.11), we would have:

fopt (S ∪ T ) ≤ f h (S ∪ T ), fopt (S ∪ T ) ≤ f h (S) + f h (T ). (2.12)

However, because we cannot predict the gap between optimal and heuristic values of the target
function, the following inequality could take place:

fopt (S ∪ T ) > f h (S) + f h (T ). (2.13)

Thus, the minimum value of the target function obtained using heuristic algorithms in the
general case does not satisfy the condition of subadditivity and cannot be used as the value of the
characteristic function.

AUTOMATION AND REMOTE CONTROL Vol. 80 No. 7 2019


COALITION-FORMATION PROBLEM 1361

To avoid this problem, it is possible to use an algorithm to correct the values of the characteristic
function DCIA, defined as follows [18]:


⎨f h (S), |S| = 1
def
c(S) = h

⎩min f (S), min(c(S\L) + c(L)) , |S| > 1.
L⊂S

Theorem (Zakharov and Shchegryaev [18]). Characteristic function calculated by DCIA algo-
rithm satisfy the condition of subadditivity.

3. ALGORITHM ALNS
ALNS (Adaptive Large Neighborhood Search)—the algorithm proposed by Ropke and
Pisinger [15]. The main idea of the algorithm is an adaptive random local search in the neigh-
borhood of the solution. The solution area here is defined as all solutions derived from applying
one of the elementary actions, such as accidentally inserting/deleting clients on a route or reassign-
ing routes. Such elementary actions are chosen randomly, according to their weights. An important
part of the algorithm is a model of simulated annealing, which makes it possible to investigate so-
lutions worse for a new search area, which allows to bypass local minima. A detailed description
of the ALNS algorithm is presented in [6].

4. DYNAMIC ADAPTATION

The idea of dynamic adaptation of heuristic algorithms is quite simple. It is based on insolvency
over time generated solutions. The goal of dynamic adaptation of a particular heuristic algorithm
is to improve the resulting solution before it is executed. Suppose that A(p) is a function of some
algorithm that compares the initial parameters of the problem p ∈ P with the generated solution.
Suppose that at each time period t = 1, 2, . . . , T − 1 we have p = pt , where pt is the state at the
time t. The state pt is the same as at the initial moment of time, but with the stock level reached
at the end of the period t and s(t, p) is the continuation of the solution that was obtained in the
previous period for the interval t + 1, t + 2, . . . , T . At the initial time, we have s0 = s(0, p): s is
the solution obtained using the A algorithm for the current subtask [19].
The general scheme of the dynamic adaptation algorithm (DA) can be represented as follows:

Algorithm 1: DA

0: Initialization: p = p0 , s = s0 , t = 1
1: while t < T do
2: s = a(p)
3: if f (s ) < f (s(t, p)) then
4: set s(k, p) = s
5: set k = k + 1
6: set p = pt

Applying dynamic adaptation to the ALNS algorithm, we obtain the DALNS algorithm. Let
ALN S(·) is a function of the ALNS algorithm, s is a solution obtained using this algorithm, S1 be
a sequence of nodes traversed by s, S2 is a sequence of remaining nodes, C is a set of problem
parameters. Then the general scheme of which can be represented by Algorithm 2.

AUTOMATION AND REMOTE CONTROL Vol. 80 No. 7 2019


1362 SHIROKIKH, LEZHNINA

Fig. 1. CIRG.

Algorithm 2: DALNS

0: Initialization: S1 = ∅, S2 = s
1: while S2 = ∅ do
2: s = ALN S(S2 , C)
3: if f (s) < f (S2 ) then S2 = s
4: C ← S2 [1], S1 ←2 [1], S2 = S2 \S2 [1]

To illustrate the algorithm, consider three different IRP [20] from the set proposed by Archetti
et al. [1], namely with high storage costs, 6 time periods and 10 clients: abs1n10, abs2n10 and
abs3n10. To build CIRG, we combined these three IRP. The resulting CIRG instance is shown
in Fig. 1. Each copy of the IRP is represented by a different set of points (squares, triangles and
circles). Each point represents the {i} depot of one carrier and j ∈ N customers.
In this game there are seven possible non-empty coalitions of carriers that can be formed. At
the first stage, we calculate the heuristic values of f h minimum costs for each coalition problem
(Table 1).
The solutions given by the ALNS and DALNS algorithms for three separate joint solutions and
full interaction are shown in Figs. 2 and 3. In the upper part of each figure there are three separate
solutions that minimize the individual costs f h (i), i = 1, 2, 3, and in the lower part—a solution that
minimizes the total costs of the coalition f h (1, 2, 3). Comparing the solutions, it is possible to see
that the DALNS algorithm gives better solutions than the ALNS algorithm.

Table 1. Heuristic estimate of the characteristic function


coalition {1} {2} {3} {1,2} {1,3} {2,3} {1,2,3}
ALNS 10 988 11 443 9866 21 135 22 062 21 567 30 335
DALNS 7885 8607 7588 15 839 15 702 15 317 22 735

AUTOMATION AND REMOTE CONTROL Vol. 80 No. 7 2019


COALITION-FORMATION PROBLEM 1363

Fig. 2. The solutions obtained by the ALNS algorithm.

Fig. 3. The solutions obtained by the DALNS algorithm.

Based on the values from the Table 1, let us define the values of the characteristic function.
Following the Direct Coalition Induction Algorithm, we take heuristic values as the characteristic
values of functions:
c({1}) = f h ({1}), c({2}) = f h (2), c({3}) = f h ({3}).
def def def

Then, we can define characteristic function for other coalitions. For coalition {1, 2}
f h ({1, 2}) = 21 135 < c({1}) + c({2}) = 22 431.

Then, we assume that


c({1, 2}) = f h ({1, 2}) = 21 135.

AUTOMATION AND REMOTE CONTROL Vol. 80 No. 7 2019


1364 SHIROKIKH, LEZHNINA

For coalition {1, 3} the following inequality holds:

f h ({1, 3}) > c({1}) + c({3}) = 20 431.

In this case, the characteristic value is assigned the following value:

c({1, 3}) = c({1}) + c({3}) = 20 854.

For coalition {2, 3} we have:

f h ({2, 3}) = 21 567 > c({2}) + c({3}) = c({2, 3}) = 21 309.

And finally, for grand coalition {1, 2, 3} we have:



⎪ h
⎨f {1, 2, 3})
⎪ = 30 335 < c({1, 2}) + c({3}) = 31 001
x= f h ({1, 2, 3}) = 30 335 < c({1}) + c({2, 3})


⎩ = c({1, 3}) + c({2}) = 32 298
⇒ c({1, 2, 3}) = f h ({1, 2, 3}) = 30 335.

For the DALNS solutions, the alterations of c(S) compare to f h (S) will be the following:

f h ({1, 2}) < c({1}) + c({2}) = 16 292 ⇒ c({1, 2}) = f h ({1, 2}) = 15 839,
f h ({2, 3}) < c({2}) + c({3}) = 15 995 ⇒ c({2, 3}) = f h ({2, 3}) = 15 316,
f h ({1, 3}) > c({1}) + c({3}) ⇒ c({1, 3}) = c({1}) + c({3}) = 15 474,
⎧ h

⎨f ({1, 2, 3}) = 22 735.3 < c({1}) + c({2, 3}) = 23 202
f h ({1, 2, 3}) = 22 735 < c({1, 2}) + c({3}) = 23 428

⎩ h
f ({1, 2, 3}) = 22 735 < c({1, 3}) + c({2}) = 23 881.

As a result, we have subadditive characteristic function in both cases. One can also note that
in this experiment, we have gotten lower value of characteristic function using DALNS.

5. STABILITY OF COALITIONS
To ensure a satisfactory level of service, each company operates its own distribution center. In
this situation, several questions may arise: is there any benefit for these firms to improve their
individual economic performance when working together? If so, what is the best coalition for each
firm, so that the profit of each cooperating firm increases? How will firms respond to collaboration
in accordance with the proposed profit sharing mechanism and investigate the incentives for a joint
recruitment situation? More specifically, the main problem is to check whether firms i ∈ I can
increase their individual profits Pi (·). At the same time, it is important to form stable coalitions:
players will not want to remain in this coalition if in the other coalition they receive greater benefits.
In this section, we take the cooperative game theory as a basis in order to propose an approach to
creating profitable coalition structures. This approach relies on the work of Aumann and Dreze [3],
which uses the concept of stability in cooperative games, taking into account the individual and
coalition rationality necessary to achieve the best compromises between all firms.
Definition 5.1. We say that a coalition S is stable among all rational coalitions of the set N if

xSi ≤ xSi for all S, S  ∈ N .

AUTOMATION AND REMOTE CONTROL Vol. 80 No. 7 2019


COALITION-FORMATION PROBLEM 1365

The purpose of the study is to find the optimal delivery schemes, as well as the search for stable
coalitions.
Profit of a player participating in a coalition depends on the chosen solution concept. In the
present paper, we investigated the stability of maximum coalitions for several current solutions
concepts.
Definition 5.2. The Shapley vector is the mathematical expectation of the contribution of each
player, if a large coalition is formed in random order
 | S |!(n− | S | −1)!  
Shi = c(S ∪ i ) − c(S) .
n!
S⊆N \ i

Definition 5.3. MSC -vector (or marginal subcore element) [18])—the principle of optimality,
which determines the player’s win according to the following formula:
⎛ ⎞
  
M SC i = ξi0 + αmsc
i
⎝c N − ξi0 ⎠ ,
j∈N
     
c S∪ i −c S
S⊆N \ i
αmsc
i =        ,
c S∪ j −c S
j∈N S⊆N \{j}
 n 
 
ξi0 : max ξi , sbj to ξi ≤ c(S), ∀S = N.
i=1 i∈S

Definition 5.4. NSCG-vector (Cost Gap Method) [9]—the optimality principle based on the
τ -value of the game and calculated by the formula:

N SCGi = mi + wi g(N ),

mi = c(N ) − c(N \ {j}), g(S) = c(S) − mj ,
j∈S
Wi
wi =  , Wi = argminS:i∈S g(S).
Wj
j∈N

Definition 5.5. EPM (Equal Profit Method) [10]—a vector that provides the highest relative
profit possible among players

EP M i = xi :
xi xj
min f : f ≥ − , ∀(i, j),
c({i}) c({j})
 
xi = c(N ), xi ≤ c(S), ∀S = N.
∈N i∈S

Considering the heuristic method of constructing the characteristic function, we cannot guaran-
tee the fulfillment of the condition of convexity of the characteristic function in the general case.
Therefore, the Shapley vector (Sh) may not lie in the C-core of the game. In the first part of
numerical experiments, similar cases were investigated. The results are presented in Table 2.
Also, a heuristic method of constructing a characteristic function can lead to a violation of
the stability of the maximum coalition. It is possible that it will be more profitable for the

AUTOMATION AND REMOTE CONTROL Vol. 80 No. 7 2019


1366 SHIROKIKH, LEZHNINA
Table 2. Location of the Shapley’s vector
The position Frequency
of the Shapley’s vector of cases
Core = ∅ 14%
Sh = Core 2.7%
Sh = Core 4.86%
Sh ∈/ Core 9%
Sh ∈ Core 69.44%

Table 3. Maximum coalition stability statistics


Number of cases Number of cases
of strict stability of strict instability
Sh MSC NSCG EPM Sh MSC NSCG EPM
Heuristic 41 28 39 41 31 44 33 25
DCIA 45 28 42 41 23 42 26 21
ALNS 46 30 42 42 25 41 29 21
DALNS 40 26 39 40 29 45 30 25
average 47 32 47 42 25 40 25 22
minimal 39 24 34 40 29 46 34 24
total 86 56 81 82 54 86 59 46
total % 59.7 39 56.3 57 37.5 59.7 41 32

player to participate in the game with a smaller number of participants. In Table 3 shows the
frequencies of strict stability (in the left part) and strict instability (in the right part) of the maxi-
mum coalition of the cooperative game in sections of various game solution concepts (by columns:
Sh, M SC, N SCG, EP M ) and various methods for solving the optimization problem (by lines).
Among the optimization medotes, a pair-wise comparison was made between simple heuristic so-
lutions and solutions using DCIA, between the ALNS and DALNS algorithms and between the
average and minimum values of the generated solutions for each problem. Additionally, total sta-
bility indicators are shown for the total set of tasks in frequency and percentage values.

6. CONCLUSIONS
Studies have shown that the stability of the maximum coalition in CIRG depends on which
heuristic algorithm was used to calculate the characteristic function. The principle of constructing
the characteristic function DCIA gives results with a higher level of stability of a large coalition
and most often leads to the non-emptiness of the core of the game. The cost allocation methods
discussed have the following order of stability of a large coalition in CIRG:

Shapley  EP M  N SCG  M SC.

REFERENCES
1. Archetti, C., Bertazzi, L., Laporte, G., and Speranza, M.G., A Branch-and-Cut Algorithm for a Vendor-
Managed Inventory-Routing Problem, Transport. Sci., 2007, vol. 41, no. 3, pp. 382–391.
2. Archetti, C., Bianchessi, N., Irnich, S., and Speranza, M.G., Formulations for an Inventory Routing
Problem, Int. Transact. Operat. Res., 2014, vol. 21, no. 3, pp. 353–374.
3. Aumann, R.J. and Dreze J.H., Cooperative Games with Coalition Structures, Int. J. Game Theory,
1974, vol. 3, no. 4, pp. 217–237
4. Bellman, R., Dynamic Programming, Princeton: Princeton Univ. Press, 1957.

AUTOMATION AND REMOTE CONTROL Vol. 80 No. 7 2019


COALITION-FORMATION PROBLEM 1367

5. Coelho, L.C., Cordeau, J.F., and Laporte, G., Thirty Years of Inventory Routing, Transport. Sci., 2013,
vol. 48, no. 1, pp. 1–19.
6. Coelho, L.C., Cordeau, J.F., and Laporte, G., The Inventory-Routing Problem with Transshipment,
Comput. Operat. Res., 2012, vol. 39, no. 11, pp. 2537–2548.
7. Cordeau, J.F., Lagana, D., Musmanno, R., and Vocaturo, F., A Decomposition-Based Heuristic for the
Multiple-Product Inventory-Routing Problem, Comput. Operat. Res., 2015, vol. 55, pp. 153–166.
8. Desaulniers, G., Rakke, J.G., and Coelho, L.C., A Branch-Price-and-Cut Algorithm for the Inventory-
Routing Problem, Transport. Sci., 2015, vol. 50, no. 3, pp. 1060–1076.
9. Driessen, T.S.H. and Tijs, S.H., The Cost Gap Method and Other Cost Allocation Methods for Multi-
purpose Water Projects, Water Resources Res., 1985, vol. 21, no. 10, pp. 1469–1475.
10. Frisk, M., Gothe-Lundgren, M., Jornsten, K., and Ronnqvist, M., Cost Allocation in Collaborative
Forest Transportation, Eur. J. Operat. Res., 2010, vol. 205, pp. 448–458.
11. Guajardo, M. and Ronnqvist, M., A Review on Cost Allocation Methods in Collabora-Tive Transporta-
tion, Int. Transact. Operat. Res., 2016, vol. 23, no. 3, pp. 371–392.
12. Jouida, J.B., Sihem, K., Krichen, S., and Klibi, W., Coalition-Formation Problem for Sourcing Contract
Design in Supply Networks, Eur. J. Operat. Res., 2017, vol. 257, no. 2, pp. 539-558.
13. Kimms, A. and Kozeletskyi, I., Core-Based Cost Allocation in the Cooperative Traveling Salesman
Problem, Eur. J. Operat. Res., 2016, vol. 248, no. 3, pp. 910–916.
14. Krajewska, M.A., Kopfer, H., Laporte, G., Ropke, S., and Zaccour, G., Horizontal Cooperation among
Freight Carriers: Request Allocation and Profit Sharing, J. Operat. Res. Soc., 2008, vol. 59, no. 11,
pp. 1483–1491.
15. Ropke, S. and Pisinger, D., An Adaptive Large Neighborhood Search Heuristic for the Pickup and
Delivery Problem with Time Windows, Transport. Sci., 2006, vol. 40, no. 4, pp. 455–472.
16. Verdonck, L., Caris, A.N., Ramaekers, K., and Janssens, G.K., Collaborative Logistics from the Per-
spective of Road Transportation Companies, Transport Rev., 2003, vol. 33, no. 6, pp. 700–719.
17. Zakharov, V. and Krylatov, A., Equilibrium Assignments in Competitive and Cooperative Traffic Flow
Routing, IFIP Adv. Inform. Commun. Techn., Springer, 2014, vol. 434, pp. 641–648.
18. Zakharov, V.V. and Shchegryaev, A.N., Stable Cooperation in Dynamic Vehicle Routing Problems,
Autom. Remote Control, 2015, vol. 76, no. 5, pp. 935–943.
19. Zakharov, V.V. and Shirokikh, V.A., Heuristic Evaluation of the Characteristic Function in the Cooper-
ative Inventory Routing Game, J. Vehicle Rout. Algorithms, 2017. https://doi.org/10.1007/s41604-017-
0003-0
20. http://www.leandro-coelho.com/instances/inventory-routing/

AUTOMATION AND REMOTE CONTROL Vol. 80 No. 7 2019

You might also like