Chapter 4 Principles and methods of model-

based decision-making in the supply chain

4.1 Basics of model-based decision-making in supply chain

management

4.1.1 Problems, systems, and decision-making

Decision-making is the major activity of supply chain managers. Each decision

implies analytical and empirical components. Analytical methods play, therefore,
a crucial role in supply chain decision-making methodologies (Kotzab et al. 2005,
Dolgui and Proth 2010, Schoenberger 2011, Render et al. 2012, Yalaoui et al.
2012, Stadtler et al. 2015). In this Chapter, we describe principles and methods of
model-based decision-making in the supply chain.
For a problem to exist there must be an individual (or a group of individuals),
referred to as the problem owner (decision maker) that:
 is dissatisfied with the current state of affairs within a real-life context, or has
unsatisfied present or future needs, i.e. has goals to be achieved or targets to be
met;
 is capable of judging when these goals, objectives, or targets have been met to
a satisfactory degree; and
 has control over some aspects of the problem situation that affect the extent to
which goals, objectives, or targets can be achieved (Daellenbach and Mc-
Nickle, 2005).
Rather than assume that we have a well-structured problem with clearly defined
objectives and alternative courses of action, it is better to:
 represent the problem situation where the issues are still vague, fuzzy
 gain a more comprehensive understanding of the various issues involved in de-
cision-making
 formulate the right problem that ensures an appropriate level of detail to
provide insights into the problem solution
 perform decision-making within the context of systems.

Most problems exist in a system context. A System is a set of interrelated entities,

comprising a whole unity where each component interacts with or is related to at
2

least one other component and they all serve a common objective. The Environ-
ment is the aggregate elements that do not belong to the system but influence it.
One of the basic system characteristics is its structure. The Structure is character-
istic of steady links and interaction modes of system’s elements. Structure de-
termines the system’s integrity, basics and building of organization and function-
ing. Structure is a framework of a system. In practice, many supply chain struc-
tures can be disrupted (e.g., Toyota’ supply chain was badly affected by the
tsunami in March 2011). Therefore, we can talk about supply chains’ structural
dynamics (Ivanov and Sokolov 2010).
Dynamics is a system’s change and evolution in the form of changes in object and
process states in space and time as driven by perturbation and control influences
of both planned nature, e.g., to transit from a current state to a desired one, and ad-
aptation nature, e.g., to adapt a system to a changed execution environment.
Decision is a selection of an activity or a set of activities to handle from several al-
ternatives. Decisions are subject to constraints, which limit decision choices and
objectives making some decisions more preferable than others do. The selection of
a managerial decision leads the system goal-oriented to the output performance.
The decisions shape the system behaviour concerning a certain individual goal or
multiple goals. Objectives and criteria play the most important role in taking de-
cisions.
Performance is a complex characteristic of the goal results of the system’s func-
tioning; potential and real performance, taking into account the conformity of the
achieved results with the goals set by management at the planning stage. Perform-
ance is measured with certain metrics or indicators (e.g., customer service level,
on-time delivery, costs, etc.)
Basic problems in decision-making include the following:
 optimality  risk and uncertainty, and
 multiple objectives  complexity
An optimal decision is the best one. Optimal decisions are very “fragile” and pre-
sume certain problem dimensionality, fullness and certainty of the model. Optimal
solutions are usually very sensitive to deviations. Moreover, the decision making
process is actually tightly interconnected with dynamics and should be considered
as an adaptive tuning process and not as a “one-way” optimization process.
Consider an example: A typical optimization problem is the determination of op-
timal order quantity for some purchased items. This optimization process is based
on the reduction of ordering and inventory holding costs. If so, the calculated or-
der quantity is optimal only for the purchasing department. At the same time, this
problem is tightly interconnected with transportation planning and customer ser-
vice level improvements. Therefore, the real problem situation, which exists, is in-
tegrated inventory optimization.
 3

The problems of applying optimization-based decision-making are interrelated

with complexity, uncertainty and multiple objectives. A particular feature of op-
timal decisions is multi-objective decision making by managers with their own
preferences that, in turn, are always changing. Hence, it becomes impossible to
build any general selection function for multi-criteria decision-making. Finding
optimal solutions is possible, but can be very time-consuming. However, these op-
timal solutions can be used for benchmarking estimations of the quality of solu-
tions obtained by heuristics or simulation models (see further in this Chapter).
Multiple objectives are conflicting in the sense of competing for common re-
sources to achieve variously different performance objectives (financial, func-
tional, environmental, etc.). A multi-criteria decision making strategy employs
trade-off analysis techniques to identify compromise designs for which the com-
peting criteria are mutually satisfied in a Pareto-optimal sense. For more detailed
information on decision-making theory, we refer to the studies on operations re-
search, decision theory, and quantitative analysis in management, e.g., Taha
(2009), Render et al. (2012), Dolgui and Proth (2010), Yalaoui et al. (2012).

4.1.2 Models and modelling

The concept of a model is widely applicable in natural human languages and is a

general scientific term. It is characterized by polysemy, that is, widely expressed
and reflecting different meanings of this concept depending on the applications
and contexts. At present, there are several hundred definitions of the concept of a
model and modelling. In summoning up different definitions, the following views
of models and modelling can be presented.

A model is:
 a system whose investigation is a tool for obtaining information about another
system;
 a method of knowledge existence; and
 a multiple system map of the original object that, together with absolutely true
content, contains conditionally true and false content, which reveals itself in the
process of its creation and practical use.
Modelling is one of the stages of cognitive activity of a subject, involving the de-
velopment (choice) of a model, conducting investigations with its help, obtaining
and analysing the results, the production of recommendations, and the estimation
of the quality of the model itself as applied to the solved problem and taking into
account specific conditions (Fig. 4.1).
4

Fig. 4.1. Modelling stakeholders

Because of the finiteness of the designed (applied) model (a limited number of ele-
ments and relations that describe objects belonging to an infinitely diverse reality)
and the limited resources (temporal, financial and material) supplied for model-
ling, the model always reflects the original object in a simplified and approximate
manner. However, the human experience testifies that these specific features of a
model are admissible and do not oppose the solution of problems that are faced by
subjects.
In the course of modelling, it is advisable to distinguish the following basic ele-
ments and relations: first, a subject or subjects, an original object, the model object
and an environment in which the modelling is performed; and, second, binary rela-
tions between the listed elements. Subjects of modelling mean the following
classes of social subjects: decision makers, persons who substantiate the decisions,
experts, persons who use the models and persons who design the models.
It is worth noting that one of the main specific features of original objects (real
or abstract) is their exceptional complexity, which reveals itself in the form of
structural complexity, complexity of functioning, complexity of the choice of be-
haviour and complexity of development. In order to describe such objects we
should use several models rather than a unique model. In other words, we should
perform system modelling (a polymodel description of the application domain).
Another specific feature of the tools of abstract modelling consists of consider-
able intensification of works in the automation of this process and, first, the phase
connected with the design of a computer model. Moreover, within the framework
of new IT based on the concepts of knowledge bases, the concept of a “model” has
considerably extended the limits of its application – from the field of passive in-
formational resources to the field of active ones. Under these conditions, al-
gorithms that are elements of procedural knowledge turn into operating environ-
ments that provide the solution to problems with a subject in the language of mod-
els.
The main properties of good models are the following:
 5

 Adequacy (from Latin adaequatus, which means equated, completely suitable,

or comparable). The model should possess the specified property relative to
certain aspects of the original object.
 Simplicity and optimality of the model. The property of adequacy is directly as-
sociated with the properties of simplicity and optimality. Indeed, sometimes, to
achieve the required degree of adequacy, we should essentially complicate the
model. On the other hand, if we can choose different models that have approx -
imately the same adequacy, it is advisable to use the simplest model.
 The flexibility (adaptability) of models assumes that parameters and structures
that can vary in given ranges are introduced into the composition of models in
order to achieve the goals of modelling.
 Universality and task orientation of models. It is advisable to design models
specialized relative to an admissible class of modelled objects and universal
with respect to a list of supported functions.
Among other properties of models, we should distinguish reliability, unifica-
tion, openness and accessibility, intelligence, the efficiency of computer imple-
mentation, complexity, identifiably, stability, sensibility, observability of models,
their invariance, self-organization and self-learning. Overall, each variant of im-
plementation of the system modelling techniques is characterized by its own time
consumption, the expenditure of resources and the results.
Finally, let us provide a general classification of different kinds of models.
There are different options for classifying models. The first and most common op-
tion is to use the morphologic analysis and to distinguish two or three possible
states for each feature. An example of such a classification is shown in Table 4.1.

Table 4.1. Morphological model classification (Ivanov and Sokolov 2010)

Kinds of mathematical
Classification feature models
I II III
Axiomatic Empirical Semi-empirical
Hypothesis deductive Deductive–asymp-
Scientific basis and building Inductive
(phenomenon logic) totic
logic
Deductive–induct-
Deductive Inductive
ive
Analytical–simula-
Exactness Analytical Simulation
tion
Data and scope Qualitative Quantitative Hybrid
Main model’s function Descriptive Predictive Hybrid
Model Selection Alternatives Satisfaction Optimization Non-alternative
Self-learning and self-organ- Strictly predeter-
Self-learning Self-organizing
ization mined
Static Kinematic Dynamic
Time
Continuous Discrete Hybrid
6

Kinds of mathematical
Classification feature models
I II III
Certainty Deterministic Stochastic Uncertainty

4.1.3 Model-based decision-making

With each decision, managers use both quantitative and qualitative models. Quant-
itative analysis is a scientific approach to managerial decision-making in which
raw data are processed and manipulated to produce meaningful information. Con-
sider an example (Ivanov et al. 2017).
Example "Re-designing the material flow in a global manufacturing net-
work"
Problem description
Consider an enterprise that produces systems for energy transmission and has
two locations: factory A is located in Europe and factory B is located in China.
Both factories have deep manufacturing penetration; in other words, they are able
to produce almost all the components and modules needed for the final product as-
sembly. Both factories can assemble the same final products from the same com-
ponents, known as shared components. The final assembly always takes place in
the country where the customer builds its energy system. It should be analysed to
see whether the production of the shared components can be distributed within the
network so that total network costs are minimized.
Model development
Each manager applies to each decision both quantitative and qualitative methods.
From the mathematical point of view, the problem can be represented as a number
of origin points with some capacities and destination points with some demand.
The production volumes should be delivered to the customers in such a way that
total costs are minimized. This is a well-known problem in operations research,
namely the transportation problem. At the same time, the problem also comprises
costs and risks, which cannot be quantified within an optimization model, and
therefore additional qualitative analysis will be needed.
Input Data
At the first stage, four options for process design have been formulated:
1. Local manufacturing at A and B
2. Manufacturing at A for A and B
3. Manufacturing at B for A and B
4. Mixed manufacturing at A and B for A and B
 7

For the development of the mathematical model, the following data is needed:
 origin points
 destination points
 production capacity at the origin points Ai
 demand at the destination points Bj
 costs for production and transportation of a product unit from Ai to Bj

Solution development
A particular feature of the problem considered is that both the origin and destina-
tion points are actually identical. The model has then the standard format of the
transportation problem and can be solved with the help of Excel Solver. Final de-
cisions should be taken based on the analysis of both quantitative optimization res-
ults and managerial qualitative analysis.
Results implementation
From the cost efficiency point of view, option 2 could be selected. However, some
crucial risks exist. First, single sourcing strategy is very risky since production and
delivery shortages can occur. Second, if manufacturing in Germany should be re-
launched, implementation costs would be very high.

The model-based decision-making process comprises therefore the following steps

as shown in Fig. 4.2.

Fig. 4.2. Model-based decision-making process (Ivanov et al. 2017)

We can observe that a real management problem is the initial point of the de-
cision-making process. An example can be a facility location problem where we
are given demand in some markets, possible locations and capacities of new facil-
ities, fixed costs for having a facility in the supply chain, and transportation costs
from each location to each market. The decision questions are where to locate the
facilities and which quantities should be shipped from the facilities to the markets.
Next step is to transform the real problem into the mathematical model. For this
transformation, we need to reduce the complexity of reality. This inevitably results
in the simplification of reality. For example, we would assume deterministic capa-
city in our facility location model instead of considering fluctuations in demand.
We do this simplification in order to represent the management problem in the
mathematical model in such a way that this model can be solved with the helped
8

of existing algorithms in a reasonable time. In our example, we formulate the fa-

cility location problem as a mixed-integer linear programming model that can be
solved with the help of simplex and branch & bound algorithms.
For implementation, software is needed. Small instances can be solved with the
help of Excel solver, but for real data, professional optimization solvers such as
CPLEX, Lindo, AMPL, Marlab, GAMS, Gurobi, XPRESS and simulation soft-
ware such as AnyLogic, anyLogistix, ARENA, LLamasoft Supply Chain Guru
and many other exist. Software will calculate the solution. In our example, it
would be the solution comprising the suggestions on where to open the facility
locations and which product quantities should be shipped from each opened loca-
tion to each of the markets so that total production and logistics costs are minimal.
Software calculated this solution. The most important question now is as follows:
is this solution automatically our decision? NO! This is a solution to the mathem-
atical problem. Now it is the stage where management expertise is needed to trans-
fer this mathematical solution into the managerial decisions. First, simplifications
of reality should be reviewed. Second, so called soft facts such as risks, flexibility,
etc. should be included in the analysis. That is why we prefer talking about de-
cision-support quantitative methods.

4.1.4 Quantitative models and Operations Research

Quantitative models for supply chain management can be divided into three
primary approaches. These are optimization, simulation and heuristics. Hybrid
models (e.g., optimization-based simulation models) also exist. Optimization is an
analysis method that determines the best possible option to solve a particular oper-
ations or supply chain problem. Optimization has been a very visible and influen -
tial topic in the field of supply chain management. The drawback of using optim-
ization is difficulty in developing a model that is sufficient detailed and accurate
in representing complexity and uncertainty of supply chain management, while
keeping the model simple enough to be solved.
Simulation is imitating the behaviour of one system with another. By making
changes to the simulated supply chain, one expects to gain understanding of the
dynamics of the physical supply chain. Simulation is an ideal tool to further ana-
lyse the performance of a proposed design derived from an optimization model.
Heuristics are intelligent rules that often lead to good, but not necessarily the best,
solutions. Heuristic approaches typically are easier to implement and require
fewer data. However, the quality of the solution is usually unknown. An option to
estimate the quality of heuristic algorithms may be the usage of optimization as a
tool for “ideal” solutions to problems.
 9

Operations Research is the application of advanced analytical methods to help

make better decisions (INFORMS definition). Operations Research deals with
problems that may be described using mathematical models to find optimal or
good solutions. Basically, Operations Research methods support the finding of
solutions to achieve the desirable goal(s) with limited resources. An overview of
Operations Research -based methods is presented in Fig. 4.3. Basic Operations
Research methods and their application to supply chain management include but
are not limited to:
 Linear programming (production planning; revenue management)
 Integer programming / Combinatorial Optimization (scheduling, routing)
 Mixed integer linear programming (supply chain design)
 Dynamic programming / Graph Theory (dynamic lot-sizing / supply chain con-
tracting)
 Network theory (project management)
 Queuing theory (waiting lines: call centres, hotline services, process design, en-
tertainment parks, services)
 Simulation (supply chain design, bullwhip-effect, etc.)
 Heuristics (genetic algorithms, ant colony optimization, application to schedul-
ing and routing problems)
 Fuzzy / Robust / Stochastic optimization (data uncertainty)

Fig. 4.3. Operations Research methods for supply chain management (Ivanov et al. 2017)

The application of many of these methods to different supply chain management

problems with structural dynamics and risk considerations will be considered in
further chapters of this book.
10

4.2 Multi-disciplinary nature of quantitative modelling

framework

Supply chain structural dynamics is characterized by uncertain interactions of the

network elements, multiple objectives, uncertain environment, and complexity.
Therefore, supply chain structural dynamics needs to be described by various
models (static and dynamic, stochastic and deterministic, analytical and simula-
tion, etc.), which are interconnected. Supply chains are also characterized by a set
of interrelated structures.
Furthermore, supply chain elements are active. Their activities are based on their
own interests and goals. Active elements necessitate balancing supply chain part-
ner interests, a large number of uncontrolled factors, and formalizing difficulties.
Besides, supply chain execution is accomplished by permanent changes in the in-
ternal network properties and the environment. It requires supply chain adaptation
to the current execution environment. So, reflections of supply chain configura-
tion, planning and execution models are needed (Ivanov 2010). The other issue of
supply chain modelling is the interlinking of conceptual and mathematical models
in order to achieve adequate, scalable and representative models and providing ap-
plication independence. Such a complex nature of supply chains requires a com-
bined application of different disciplines and modelling approaches (Choi et al.
2016). Fig. 4.4 depicts supply chain integrated modelling framework.
 11

Fig. 4.4 Supply chain integrated modelling (based on Ivanov 2009)

The main parts of the multi-disciplinary methodology are as follows: the gen-
eral systems framework, the integrated modelling framework, and the computing
framework (Fig. 4.5).
12

Fig. 4.5 Integrated modelling “concept-model-computation” (Ivanov 2009)

The general system framework defines conceptualized business models, meta-

methodologies and a set of categories, definitions, specifications and performance
indicators, which are developed during the integration of various theoretical
frameworks. The integrated modelling framework defines the rules of the integ-
rated multi-disciplinary mathematical model building. It proposes constructive
methods and techniques of (1) how to combine various model classes and (2) how
to interconnect partial supply chain problems. The computing framework integ-
rates the building of mathematical models and algorithms, and their implementa-
tion as software.
Model integration as a key element of multi-disciplinary modelling framework
is considered from four perspectives of system modelling: the integration of vari-
ous modelling approaches and frameworks, the integration of planning and execu-
tion models, the integration of decision-making levels, and the implementation of
integration throughout: “conceptual model  mathematical model  computa-
tion” (see Figs. 4.6 and 4.7).
 13

Fig. 4.6 Mathematical integration view 3

Fig. 4.7 Decision-making integration view

Supply chain strategy, design, planning and operations are interlinked with re-
gards to constructing ideal supply chain states as well as to reconfiguring supply
chains in relation to a current execution environment at strategic, tactical, and op-
erational levels. The large variety of supply chain management issues can be clas-
sified into the subclasses of supply chain analysis and synthesis. The models of
supply chain analysis can be divided into supply chain design analysis as well as
supply chain operative monitoring (see Fig. 4.8).
14

Fig. 4.8 Complex of conceptualized models for the problem of the supply chain dynamical struc-
tural-functional synthesis and reconfiguration (adopted with changes from Ivanov 2009)

The models of supply chain synthesis are composed of supply chain configura-
tion and reconfiguration models. Most supply chain problems are cross-linked (i.e.
models of supply chain monitoring and reconfiguration, static and dynamic mod-
els of the supply chain (re)configuration, supply chain synthesis and analysis mod-
els). Conceptual, mathematical and information models are also interconnected
with each other.
The supply chains need to be configured according to project goals and recon-
figured in dynamics according to the current execution environment. More typic-
ally, questions are centred on rationalizing supply chains in response to permanent
changes in the supply chain itself and its environment. The general modelling
structure of the supply chain dynamical structural–functional synthesis and recon-
figuration is shown in Fig. 4.9.
Network design Network
control
Definition of Definition of active Description of
structures and their elements and their network operation
interconnections interractions (semi- processes Adaptive Supply chain
(static network dynamic network (dynamic network planning monitoring:
description) description) description) and control - measurement,
- diagnosis,
- forecasting.
Graph Multi-Agent Dynamic Plan
Model System System state

Multi-Agent conceptual
Multiple-Model complex
modeling framework Actual
state Supply chain
regulation
(reconfiguration)
Definition of goals Description of Description of
criteria, forms of uncertainties, risk scheduling
their representation factors, and methods algortihms
and evaluation of their control

Fig. 4.9 The general modelling schema of the supply chain dynamical structural-functional syn-
thesis and reconfiguration (Ivanov et al. 2007)
 15

Let us consider some examples. Transportation and inventory are primary com-
ponents of the order fulfilment process in terms of cost and service levels. There-
fore, companies must consider the important interrelationships among transporta-
tion, inventory and customer service in determining their policies. The suppliers’
selection is linked not only to their capacities, costs, etc. but also to their collabor-
ation abilities. Therefore, coordination between the various players in the chain is
the key to its effective management. Pricing and inventory decisions as well as
product, distribution and production decisions are also matched together.
Let us consider the main steps of this schema referring also to the main phases
of the supply chain management in Figs 9.5-9.6. The modelling starts with the
static graph-theoretical network description. The supply chain can be described
graph-theoretically as a directed graph — a digraph. Then the elements of the or-
ganizational graph (enterprises) are described as active agents in terms of multi-
agent theory. So the model of enterprise interactions can be constructed. In this
stage, we combine graph-theoretical modelling with active agents to describe act-
ive elements of the graph as well as to implement modelling dynamics of the sup-
ply chain objects’ collaboration.
The goal of the coordinator consists of project description and structuring ac-
cording to the proper level of decomposition. Any project can be presented as sev-
eral consecutive and/or parallel operations. In complex projects, different opera-
tions have different degrees of importance to the final result. That is why the im-
portance level of each operation must be taken into account. When a coordinator is
in charge of several projects, there may be situations where the same competency
is needed for different projects at the same time. These projects should not be con-
sidered independently because they compete for the same resources. In order to
analyse such “linked” projects, we propose to present them as a joined structure
scheme that we call a technological network. The model of the technological net-
work is an oriented graph. Its heads are considered competencies that are neces-
sary for project realization, and its edges serve to show the logical sequence of op-
erations. The weights of each head reflect the volume of the competency that is
needed for the project’s performance (e.g. the total working time to execute a con-
crete technological operation on a concrete machine). Finally, at the end of this
step, the coordinator obtains properly structured projects that are considered as a
technological network. The next step is dynamic network description. In this
stage, a set of interlinked dynamic models is formed to link the “ideal” planning
results to the supply chain execution programmes under uncertainty.
The models of supply chain execution are comprised of supply chain monitor-
ing and supply chain regulation (reconfiguration) models. Supply chain monitor-
ing is based on the monitoring of the supply chain macro-structural macro-states
(cf Chapter 1). The monitoring plan execution consists of determining diagnostics
moments (critical control points), when the analysis of the planned and factual
parameters of supply chain execution (demand, inventories, jobs’ starting and end-
ing, and stability) is carried out. A particular feature of supply chain monitoring in
terms of macro-states is that, at each monitoring stage, the control parameters are
16

extracted from the parameter vector of the dynamic alternative multi-graph. The
mathematical description of the dynamic alternative multi-graph has been con-
sidered in Chapter 1, see Eq. (1.1). The extracting rules depend on management
goals at the stage monitored. This makes it possible to consider all of the supply
chain execution parameters described in the dynamic alternative multi-graph and
to extract the necessary control parameters in the current execution situation.
Supply chain reconfiguration (real-time re-planning) is comprised of deviations
analysis, the elaboration of compensating control actions and the construction of a
new plan and production of appropriate correcting actions for the transition from
the actual supply chain state trajectory to the planned one at a given time interval
or by the final time. The model of supply chain reconfiguration is interconnected
to the planning model. It is also based on the dynamical alternative multi-graph.
While selecting a new supply chain, it is also essential to take into account a num -
ber of specific requirements (preferences of the supply chain focal enterprise or
4PL provider and suppliers at different supply chain levels). The essence of de-
cision making about supply chain reconfiguration, that is to say about a supply
chain plan transition from a current state to a desired state, is to ensure that the
agents’ interests and the interests of a supply chain coordinator comply with each
other.

4.3. Modelling paradigms

4.3.1. Mathematical optimization

Optimization is an analysis method that determines the best possible method of

designing a particular supply chain. Optimization methods have been a very vis-
ible and influential topic in the field of supply chain management (Yalaoui et al.
2012, Martel and Klibi 2016). The main advantage of the optimization approach is
the idea of optimality and striving for the best solution. However, the optimal ap-
proach should be very carefully considered. The problems of applying optimiza-
tion-based decision-making are tightly interrelated with complexity, uncertainty
and multiple objectives. Optimal approaches are very “fragile” and presume cer-
tain problem dimensionality, fullness and certainty of the model. Besides, optimal
solutions are usually very sensitive to deviations. Moreover, the decision making
is actually tightly interconnected with dynamics and should be considered as an
adaptive tuning process. A particular feature of complexity in supply chains is
multi-criteria decision making by managers with their own preferences that, in
 17

turn, are always changing. Hence, it becomes impossible to build any general se-
lection function for multi-criteria decision-making.
The drawback of using optimization is the level of difficulty in developing a
model that is sufficient detailed and accurate in representing complexity and un-
certainty of supply chain management while keeping the model simple enough to
be solved. Furthermore, most of the models in this category are deterministic and
static. Additionally, those that consider stochastic elements are very restrictive in
nature.
Indeed, with regards to supply chains as complex systems, optimization can be
realistically considered as the direction for best solutions and the ideology of de-
cision-making. Finding optimal solutions is possible, but it can be very time-con-
suming. However, these optimal solutions can be used for benchmarking to estim-
ate the quality of solutions obtained by heuristics or simulation models. Unless
mitigating circumstances exist, optimization is the preferred approach for supply
chain management. However, in reality, only a few partial supply chain manage-
ment problems (mostly of a strategic nature) may be correctly addressed by optim-
ization.

4.3.2. Simulation

Equation systems that are good at expressing static dependencies between para-
meters and variables typically are restrictive in describing systems with dynamic
behaviour. It’s why another modelling technology - simulation modelling - to ana-
lyse dynamic systems is applied. Simulation involves imitating the behaviour of
one system with another. By making changes to the simulated supply chain, one
expects to gain understanding of the dynamics of the physical supply chain.
Three major simulation approaches have be identified (Tako and Robinson
2012, Bortchev 2013):
 discrete-event simulation
 system dynamics
 agent-based simulation
Each method serves a specific range of abstraction levels. System dynamics as-
sumes very high abstraction, and it’s typically used for strategic modelling. Agent
based models can vary from very detailed models where agents represent physical
objects to highly abstract models where agents represent competing companies or
governments. Discrete event modelling supports medium and medium-low ab-
straction.
Supply chain simulation utilizes three classes of software tools:
 all-purpose discrete event simulation tools;
 specific supply chain simulation software; and
 supply chain management software with simulation functionality.
18

Two classes of dynamic models, namely continuous and discrete models, are
widely used to depict process sequences in flow systems. Continuous models are
based on differential equations and are most frequently applied as system dynam-
ics models to reproduce manufacturing and logistics processes (Sterman 2000).
System dynamics models are relatively rarely developed since planners often find
the models in this class too abstract or are unable to generate them. Since these
models normally relate relatively rough to real problems and very abstractly, they
are hereafter referred to as macroscopic models (Tolujew 2008). Operative plan-
ning routinely applies microscopic discrete-event models. The principles and tools
of discrete-event simulation are utilized to implement discrete models. Since
workstations, technical resources, carriers and units of goods are represented as in-
dividual objects in most cases, event-oriented models may also be referred to as
microscopic models.

4.3.3. Optimization-based simulation

One promising area in model-based decision-making is the study of combining

simulation methods with optimization methods in an iterative way that is referred
to as an optimization-based simulation. For example, Meisel and Bierwirth (2014)
utilized simulation and optimization methods to analyse the performance of a
make-to-order strategy in the presence of uncertainties. Ivanov et al. (2016) ap-
plied a hybrid optimization-based simulation approach to analysing supply chain
disruptions and recovery strategies.
In Table 4.2, advantages and limitation of optimization are summarized.

Table 4.2. Advantages and limitations of optimization

Advantages Possible limitations

Optimal solution Dimensionality and complexity of real problems
Dynamics and uncertainty of system and model evolu-
Methodical basics
tion
Easy accessibility Poor flexibility
Linearity and discreteness

Simulation methods do not guarantee optimality and development of simulation

model can be a complex task even with the use of special software. However, sim-
ulation provides a number of advantages that compensate for optimization limita-
tions. In particular, in simulation models it has become possible to
 analyse systems and find solutions where methods such as analytic calculations
and mathematical programming fail
 measure values and track entities within the level of abstraction
 19

 add measurements and statistical analysis any time

 manipulate and animate the system behaviour in real time
Therefore, simulation is an ideal tool for further analysing the performance of a
proposed design derived from an optimization model (Ivanov 2017).

4.3.4. Control theory

Control theory (control theory) as a base for studying multi-stage, multi-period

dynamic systems is an interesting research avenue to extend existing results while
taking into account the intrinsic peculiarities of modern supply chains. Control
theory contains a rigor quantitative basis for planning optimal control policies in-
cluding differential games and stochastic systems, stability of controlled processes
and non-linear systems, controllability and observability, and adaptation (Perea et
al., 2000, Sethi and Thompson 2000, Disney et al., 2006; Sarimveis et al., 2008,
Ivanov et al. 2012, Fu et al. 2015).
These tools can be applied for a wide range of systems, from discrete linear to
stochastic non-linear systems with both stable and dynamically changing struc-
tures. Control theory can also be applied to analysis of equilibriums of resource
consumption and system output. Applications of control theory, simulation and
systems science to supply chain dynamics are multi-faceted. Disney et al. (2006),
Sarimveis et al. (2008) and Ivanov et al. (2012) underline the resemblance of sup-
ply chains to engineering dynamic systems.
Control theory is a multi-disciplinary scientific discipline that contains power-
ful conceptual and constructive tools to conduct research on the dynamic problems
of flexible (re)distribution of a variable set of jobs to a variable set of resources.
Control theory is tightly interconnected with cybernetics. Wiener (1948) pro-
posed summing up the whole area of regulation and communication theory under
the term “cybernetics”: “We have decided to call the entire field of control and
communication theory, whether in the machine or in the animal, by the name Cy-
bernetics”.
The renovation of cybernetics has two sources. The first source lies in the at-
tempts to revise the methodological backgrounds of cybernetics. Maruyama
(1963) paid attention to the systems in which mutual causal effects are deviation-
amplifying. Economic, social, and biological examples were considered.
In contrast to Weiner’s cybernetics with deviation-counteracting systems, the
studies of deviation-amplifying mutual causal relationships were called “the
second cybernetics”. Von Foester (1974) defined “the second-order cybernetics”
with awareness that an observer is an element of the system. The studies con-
sidered processes resulting in an increase in biological and social complexity.
Stafford Bear, in his work since 1974, has emphasized that investigation into com-
plexity problems should evolve Ashby’s law of requisite variety (Ashby 1956).
20

Unfortunately, the logically relevant chain of fundamental notions of cybernet-

ics – control – informational processes – universal transformer of information
(computer, cybernetic machine) was split. An expansion of computer technologies
caused the illusion of their ability to solve any problem. The imperfection of these
technologies has already caused catastrophes that allowed scientists to proclaim
the establishment of a “risk society” rather than an “informational” one. This in-
spires a renewed interest in the theoretical background of control problems.
Within control theory, different control approaches exist, i.e. optimal control,
adaptive control, and intelligent control. One of the most popular techniques of
optimal control is the model predictive control. Model predictive control has been
a preferred algorithm for robust, multi-variable control that has been widely used
in process industries. The popularity of Model predictive control stems from the
relative ease with which it can be understood, and its ability to handle input and
output constraints.
Model predictive control is a control strategy based upon the explicit use of a
process model to predict the process output (performance) over a long period of
time (Camacho and Bordons 2004). The model attempts to predict the control
variables for a set of time periods. Predicted control variables depend on disturb-
ance forecasts (i.e. demand, prices and interest rates) and also on a set of given
parameters that are referred to in control literature as control inputs.
In Model predictive control, a system model and current and historical meas-
urements of the process are used to predict system behaviour at future time in -
stants. A control-relevant objective function is then optimized to calculate a se-
quence of future control moves that must satisfy the system constraints. As a con -
trol-oriented framework, a model predictive control-based planning scheme has
the advantage that it can be fine-tuned to provide acceptable performance in the
presence of significant uncertainty, forecast error, and constraints on inventory
levels, production, and shipping capacity. However, the “optimal control” struc-
ture in model predictive control is only a means to achieve such a result as it does
not optimize a true performance index of the closed-loop control system.
Adaptive control is a control strategy with some form of recursive system iden-
tification. Usually, a parametric adaptive control is considered (Sastry and Bodson
1994). Research in AC has a long and vigorous history. Kalman (1960) developed
the concept of a general self-tuning (adjusting) controller with explicit identifica-
tion of the parameters of a linear, single-input, single-output plant and proposed to
use these parameters’ estimation to update an optimal linear quadratic controller.
In the 1960–70s, due to the establishment of Lyapunov’s stability and proving
convergence in adaptive control systems, stochastic control made giant strides un-
derstanding dynamic programming due to Bellmann and others (Bellmann 1972).
In the 1980s, adaptive schemes for different applied domains appeared (Goodwin
et al. 1980). However, the adaptive control approach has not found a wide applica-
tion in the supply chain management domain. The main cause of this is that adapt -
ive control techniques are intended for technical systems with an automatic con-
troller and automatic reactions of milliseconds. This is not the case in supply
 21

chains. However, the main principles of adaptive control (not the formal tech-
niques) can enrich the control framework for supply chain management that will
be shown later in this study.
Interest in Control theory is of great interest to researchers and practitioners.
Control theory is favourable in cases of many dynamically changing control para-
meters, obtaining analytical solutions or properties, and in investigating different
mutual impacts of supply chain planning and control parameters (e.g., demands,
resource and channel capacities, lead-time, lot-sizes, and inventories) in the supply
chain tactical and operative performance (i.e., service level and costs). In some
cases (e.g., if there are many changes, many stages, and many periods), it is con -
venient to transit from a discrete problem statement to continuous solution proced-
ure, and then represent the result again in discrete terms due to particular accuracy
of continuous time models.
Employment of control theory can be seen as the next crucial step in the devel-
opment of supply chain management theory to reflect real-time dynamics and dy-
namic optimization of supply chain structures and processes as well as explore ro-
bustness, stability, and adaptability in real-time mode taking into account non-lin-
earity, non-stationarity and uncertainty in supply chains.
Dynamics in supply chains can be referred to both as the dynamics of a process
under optimization (dynamics of the transition from an input to an output state)
and the real-time dynamics regarding the feedback-loop consideration and adapta-
tion in accordance with an actual execution environment. Employment of control
theory can be favourable for both supply chain synthesis and analysis stages.
Optimal control approaches can be efficiently used for optimal design, planning
and scheduling of complex multi-stage, multi-period and multi-commodity supply
chains. However, an important precondition is the existence of centralized inform-
ation for the whole network and relevant parameters. In practice, it is rather rare
that firms really share complete data on inventories, demands, and capacities.
Therefore, optimal control can be efficiently applied only for centralized informa-
tion strategies.
Such cases are very common in supply chains, e.g., with original equipment
manufacturers, by applying the CPFR (collaborative planning, forecasting and re-
plenishment) coordination strategy or in supply chains which are managed by lo-
gistics service providers. However, even if decentralization and contradictory
goals of enterprises in a supply chain do not allow for the implementation of op-
timal program control, optimal solutions under assumptions of full information
sharing may be considered as an orientation for the estimation of any decentral-
ized plans and schedules that can be found, e.g., with the help of agents.
Linear control theory can be efficiently applied to multi-echelon inventory con-
trol problems with full information sharing, e.g. vendor-managed inventory. In ad-
dition, information delays can be efficiently investigated with linear control the-
ory. The bullwhip-effect and stability of supply chains can be approached in the
frequency domain of the linear control theory, though purely under the assumption
of relatively small dimensionality. For the stage of execution control (again, under
22

the assumption of immediate and full information sharing), model predictive con-
trol methods can also be applied.
The following research issues can be indicated:

 Issue 1 – Synthesis models - complexity and content of planning models and

their interconnection with control models. Supply chains are increasingly more
and more complex. It is becoming ever more difficult to represent the progress-
ively more complex supply chain management problems within only one model
taking into account dynamic supply chain processes.
 Issue 2 – Analysis models - executability, robustness, stability and adaptation.
Planned performance success can be thwarted by perturbations. This forces re-
search on re-planning and control to make supply chains reliable and flexible
enough to be able to adapt when disturbed and remain stable and robust.
With regard to the abovementioned research gaps, employing control theory
methods is becoming timely and crucial. Supply chains resemble control systems
as multi-stage dynamic flow of materials with information feedback. Control the-
ory contains a rigorous quantitative basis for planning multi-stage dynamic sys-
tems. Control theory tools can be applied to a wide range of systems, from discrete
linear to stochastic non-linear systems with both stable and dynamically changing
structures. In Table 4.3, possible applications of modern control theory results to
supply chain management are summarized.

Table 4.3 - Applications of modern control theory to supply chain management (Ivanov et al.
2012)
The main results of modern control Implementations for supply chain control
theory
Criteria for existence of a solution Model verification for supply chain control

Criteria for controllability and attain- Control processes verification for a given time interval /
ability Determination of the constraints on supply chain goals

Criteria for uniqueness of optimal Analysis of possibility to obtain an optimal plan

program control
Necessary and sufficient conditions Preliminary analysis of optimal program controls; genera-
of optimality tion of basic supply chain planning algorithms

The program control and feedback supply chain planning, scheduling and execution control
control models on united methodological basis

Criteria for stability and resilience Evaluation of supply chain robustness, stability and resili-
ence for environmental impacts and alteration of input data
 23

Control theory takes into account dynamics, real dimensions, non-linearity and
non-stationary of supply chain processes. However, although supply chains re-
semble control systems, they have some peculiarities which do not allow direct ap-
plication of control theory methods. Classical control theory leads us to the fields
of automatic control. In supply chains, the controllers are human beings.
Supply chain fine tuning occurs not within milliseconds but with a time delay.
Supply chain managers consciously tend to take risks. In light of multi-criteria
problems, decisions in supply chains are typically made based on individual psy-
chological risk perceptions and preferences. Hence, interactive tools for multi-cri-
teria decision-making are needed. Decision-making in supply chains is of a dis-
crete nature. In technical control systems, it is assumed that control is selected
continuously. Besides, in optimal control theory, differential equations express the
process dynamics of the systems’ behaviour in the input-output context. These
non-linear mathematics in a supply chain management context are a promising re-
search area (Spiegler et al. 2016).
We conclude that one of the fascinating features of control theory is the ex-
traordinarily wide range of possible applications. Control theory is vital in unlock-
ing many system improvement opportunities. The first strong contribution of con-
trol theory to operations and supply chain management regarding the dynamics is
the interpretation of planning and execution processes not as isolated domains but
as an adaptive process. Second, an advantage of control theory is the possibility to
solve problems with non-stationary and non-linear processes due to the independ-
ency of time variables. Continuous dynamic models allow us to establish and op-
timize supply chain performance indicators in dynamics that are difficult to ex-
press within static and discrete time models. Third, the possibility of covering sup-
ply chain dynamics at the process level and the changes in the supply chain and
the environment are also strong contributions of control theory. Fourth, control
theory allows for the consideration of goal-oriented formation of supply chain
structures and the solution of problems in this system as a whole. Fifth, in using
control theory, important categories of supply chain analysis such as stability, ro-
bustness and adaptability can be taken into consideration.
However, some peculiarities of supply chains as technical-organizational sys-
tems do not allow direct application of control theory methods. For this reason, a
combined application of control theory, operations research, and artificial intelli-
gence can be favourable. In addition, ipso facto that a supply chain plan and
schedule can be formulated as optimal program control is a great advantage sub-
ject to further dynamics analysis that is a crucial application area of control theory
to supply chain management. Even in this area, control theory can be applied to
supply chain management to a greater extent and enlarge the scope of supply
chain dynamics analyses that is currently rather limited.
24

4.3.5. Heuristics

Heuristics are intelligent rules that often lead to good, but not necessarily the best,
solutions. Heuristic approaches typically are easier to implement and require less
data. However, the quality of the solution is usually unknown. Unless there is a
reason not to use optimization, heuristics are an inferior approach. In supply chain
settings, nature-based heuristics such as genetic algorithms and Ant Colony Op-
timization are usually applied.
Heuristics (e.g. genetic algorithms) are usually applied instead of optimization
because of computational complexity. They do not guarantee the optimal solution
but allow a permissible result to be found within an acceptable period of time. The
quality of this solution with regards to the potential optimum, however, remains
unknown. Second, multiple objective problems are still a “bottleneck” of heurist-
ics.

References

Ashby WR (1956) An introduction to cybernetics. Chapman & Hall, London

Bellmann R (1972) Adaptive control processes: a guided tour. Princeton Univ. Press: Princeton,
New Jersey
Borshev A (2013) The big book of simulation modelling. AnyLogic North America
Bramer M (2013) Principles of data mining. 2nd edn. Springer, London
Camacho EF, Bordons C (2004) Model predictive control. Springer, London
Casti JL (1979) Connectivity, complexity and catastrophe in large-scale systems. Wiley-Inter-
science: New York, London
Choi TM, Cheng TCE, Zhao X (2016) Multi-Methodological Research in Operations Manage-
ment. Prod Oper Manag 25:379-389
Daellenbach HG, McNickle DC (2005) Management Science. Decision making through systems
thinking. Palgrave Macmillan, Basingstoke
Disney SM, Towill DR, Warburton RDH (2006) On the equivalence of control theoretic, differ-
ential and difference equation approaches to modelling supply chains. Int J Prod Econ
101:194-208
Dolgui A Proth JM (2010) Supply Chain Engineering: Useful Methods and Techniques.
Springer, Berlin
Foerster von H (1974) Cybernetics of Cybernetics. Paper delivered at 1970 annual meeting of the
American Society for Cybernetics. University of Illinois, Urbana
Fu D, Ionescu C, Aghezzaf EH, De Keyser R (2015) Quantifying and mitigating the bullwhip ef-
fect in a benchmark supply chain system by an extended prediction self-adaptive control or -
dering policy. Comput Ind Eng 81:46-57
Heizer J, Render B (2013) Operations Management: Sustainability and Supply Chain Manage-
ment. 11th edn. Pearson, Harlow
Ivanov D (2009) DIMA - a research methodology for comprehensive multi-disciplinary model -
ling of production and logistics networks. Int J Prod Res 47(5):1133-1155
Ivanov D (2010) A framework for aligning (re)planning decisions on supply chains strategy,
design, tactics, and operations. Int J Prod Res 48(13):3999-4017
 25

Ivanov D, Kaeschel J, Sokolov B (2007) Integrated modelling of agile enterprise networks. Inter-
national Journal of Agile Systems and Management 2(1):23-49
Ivanov D, Sokolov B, Pavlov A, Dolgui A, Pavlov D (2016) Disruption-driven supply chain (re)-
planning and performance impact assessment with consideration of pro-active and recovery
policies. Transport Res E-Log 90:7-24
Ivanov D, Tsipoulanidis A, Schönberger J (2017) Global Supply Chain and Operations Manage-
ment: A decision-oriented introduction into the creation of value. 1st edn. Springer
Ivanov D, Sokolov B (2010) Adaptive Supply Chain Management. Springer, London et. al.
Ivanov D, Sokolov B, Dolgui A (2012) On applicability of control theory to adaptive supply
chain planning and scheduling. Annu Rev Control 36(1):73-84
Ivanov D. (2017) Supply Chain Analytics with anyLogistix. E-book. Available at blog.hwr-ber-
lin.de/ivanov
Kalman RE (1960) A new approach to linear filtering and prediction problems. J Basic Eng
82(1):35-45
Kotzab H, Seuring S, Müller M, Reiner (2005) Research Methodologies in Supply Chain Man -
agement. Physica-Verlag, Heidelberg
Martel A, Klibi W (2016) Designing Value-Creating Supply Chain Networks. 1st edn. Springer
Maruyama M (1963) The second cybernetics. Deviation amplifying mutual causal process. Am
Sci 51:164-179
Meisel F, Bierwirth C (2014) The design of make-to-order supply networks under uncertainties
using simulation and optimization. Int J Prod Res 52(22):6590-6607
Mesarovic MD, Takahara Y (1975) General systems theory: mathematical foundations. Aca-
demic Press: New York, San Francisco, London
Perea E, Grossmann I, Ydstie E, Tahmassebi T (2000) Dynamic modelling and classical control
theory for supply chain management. Comput Chem Eng 24:1143-1149
Render B, Stair RM, Hanna ME (2012) Quantitative analysis for management. 10th edn. Pear-
son, Harlow
Sarimveis H, Patrinos P, Tarantilis CD, Kiranoudis CT (2008) Dynamic modelling and control of
supply chain systems: A review. Comput Oper Res 35:3530-3561
Sastry S, Bodson M (1994) Adaptive control: stability, convergence, and robustness. Prentice-
Hall, New Jersey
Schönberger J (2011) Model-Based Control of Logistics Processes in Volatile Environments:
Decision Support for Operations Planning in Supply Consortia. Springer, New York
Sethi SP, Thompson GL (2000) Optimal Control Theory: Applications to Management Science
and Economics. 2nd Edn. Springer, Berlin
Spiegler VLM, Potter AT, Naim MM, Towill DR (2016) The value of nonlinear control theory in
investigating the underlying dynamics and resilience of a grocery supply chain. Int J Prod
Res 54(1):265-286
Stadtler H, Kilger C, Meyr H (2015) Supply Chain Management and Advanced Planning. 5th
edn. Springer, Berlin et al.
Sterman J (2000) Business dynamics. McGraw-Hill Education Ltd.
Taha HA (2009) Operations Research: An Introduction. 9th edn. Pearson, Harlow
Tako AA, Robinson S (2012) The application of discrete event simulation and system dynamics
in the logistics and supply chain context. Decis Support Syst 52:802-815
Tolujev J (2008) Practical issues in simulation modelling logistics systems. Logistics and Supply
Chain Management 25(2):53-63 (in Russian)
Wiener N (1948) Cybernetics or control and communication in the animal and the machine. John
Wiley & Sons Inc, New York
Yalaoui A, Chehade H, Yalaoui F, Amodeo L (2012) Optimization of Logistics. Wiley, New
York
26

A multi-objective decision making 3

active elements 10 O
C Operations Research 9
control optimality 2
adaptive 19 optimization 15
Control theory 18 Optimization 8
cybernetics 18 optimization-based simulation 17
D P
Decision 2 Pareto-optimal 3
dynamics Performanc 2
system 17 problem 1
Dynamics 2 S
E simulation 8, 16
effectiveness 5 discrete-event 17
efficiency 5 structural dynamics 2
H supply chain 10
heuristics 8 analysis 12
modelling 10
M models 10
model 3 monitoring 14
classification 5 reconfiguration 15
continuous 17 simulation 16
dynamic 17 synthesis 12
integration 11 System 1
model predictive control 19 system dynamics 17
modelling 3
adequacy 5
T
flexibility 5 trade-off 3