=Paper=
{{Paper
|id=Vol-2922/paper017
|storemode=property
|title=Situations representation and retrieve in the case-based reasoning system for managing a complex technological object
|pdfUrl=https://ceur-ws.org/Vol-2922/paper017.pdf
|volume=Vol-2922
|authors=Igor Glukhikh,Dmitry Glukhikh
}}
==Situations representation and retrieve in the case-based reasoning system for managing a complex technological object==
https://ceur-ws.org/Vol-2922/paper017.pdf
Situations representation and retrieve in the case-based
reasoning system for managing a complex technological
object*
Igor Glukhikh and Dmitry Glukhikh
University of Tyumen, 6 Volodarskogo Street, Tyumen, 625003, Russian Federation
gluhihdmitry@gmail.com
Abstract. Modern urban infrastructure systems are complex
technological objects. Their stable operation is important for facilitating
a comfortable and safe urban environment. These systems are supported
by monitoring and quickly addressing potentially dangerous situations.
Due to the high complexity and high level of responsibility of decision-
making in dangerous situations, the problems of intelligent decision-
making support are relevant. The article explores the use of the case
based reasoning (CBR) method for solving these problems. In CBR, the
knowledge base of a decision support system contains cases: situations
and solutions that are applicable in such situations. When a dangerous
situation arises, the system turns to the knowledge base to search for a
case with a prepared solution. For realization of the CBR method, the
paper proposes a situation is represented through the states of the
elements of a complex object and the relationships between them. For
the retrieve of cases in the knowledge base, an approach that takes into
account the structural and parametric similarity of situations is proposed.
Keywords: case-based reasoning, intelligence monitoring, decision
support systems, urban infrastructure.
1. Introduction
Modern urban infrastructure systems (electricity, gas, water supply, and heating) are
complex technological objects (CTO). The safety and stability of their processes are
important not only for supporting the life in the city, but also for the preservation of the
environment, health and lives of people.
These systems are supported by monitoring their condition and prompt
troubleshooting. The tasks of monitoring complex objects in order to prevent
* Copyright c 2021 for this paper by its authors. Use permitted under Creative Commons License Attribution
4.0 International (CC BY 4.0).
emergencies are relevant for enterprises supplying heat, water, gas, and energy to the
region, as well as for security and urban management services.
Quite a lot of recent papers are dedicated to the research in the field of monitoring
of technological objects. Primary data collection and processing are a priority, and
various technical methods, devices and communication channels are being developed
for this [1-4]. The next big task is to analyze the data and predict the dynamics of
changes in the condition of the object and to identify emergency situations. For this,
methods of data mining, as well as machine learning and artificial neural networks are
used [5-12]. However, after identifying dangerous situations, a complex of tasks at the
following levels appears. These are the tasks of decision -making in dangerous
situations and preventing their cons equences, as well as the tasks of analyzing,
identifying and addressing the causes of such situations in order to prevent them from
occurring in the future. This paper is dedicated to the tasks of the level of decision-
making support in the prompt elimination of dangerous situations. The case based
reasoning (CBR) method known from artificial intelligence studies is considered as a
base for this.
The CBR method involves maintaining a knowledge base (KB) where cases,
descriptions of complex situations known from past experience and solutions that have
been recommended or used in these situations before, are stored. When a new
dangerous situation arises, the KB contains a case with the same or a similar situation
and a solution that is provided to users. The s olutions found in this way can be used
directly or adapted to the situation at hand.
Case based reasoning is widely used in different subject areas. One of the promising
areas is associated with decision-making when managing complex technical and
organizational objects [13-16]. At the same time, due to the complexity and diversity
of the objects under consideration, each problem area still requires its own research,
starting with the search for models for formalizing the representation of objects and
continuing with algorithms for inference and adaptation of solutions.
The goal of this research is to develop generalized (universal) models of
representation of situations for the application of the case based reasoning method when
making decisions in dangerous situations on complex technological objects of urban
infrastructure. The article first describes the content and stages of the CBR method,
then develops a general ontological model (GOM) of a complex object. Then, based on
this model, a formalized representation of situations arising at a technological object is
developed. Further, an approach and methods for assessing the similarity of situations
are proposed, taking into account both the parameters and the structure of situations on
a complex object. After that, the results are discussed and tasks for further research are
proposed.
2. Materials and methods
The case based reasoning method was widely used in different subject areas. In the 80s
and 90s of the last century, a number of commercial systems were developed using
CBR (CLAVIER, CHEF, CASEY, JULIA, etc.) that convincingly demonstrated the
effectiveness of this method [17]. Today, the development of CBR is associated, on the
one hand, with the expansion of the scope of applications and the development of
practical applications; on the other hand, with the research that will aid in combining
the methods of presentation of knowledge and machine learning for an integrated,
neurosymbolic approach to the creation of artificial intelligence [18-20]. In a CBR
system, a knowledge base is a case base, each of which is a pair , where Sit
is a situation that required its solution; Sol is the solution for this situation. As a
solution, one can write down what has already been applied in practice in that particular
situation and has shown positive results. Or it can be a solution that was specially
developed by experts in advance for such situations.
The main stages of inference in the CBR system are:
1. Identifying the actual situation Sit act;
2. Retrieve the Sit* situation, which is closest to the current situation Sit act.
Applying a solution from a pair of < Sit* , Sol*> to resolve the Sit act. In this case, if
the situation Sit* is not close enough to the Sit act, the solution Sol* acts as a basis for
generating the adapted Solact Sitact.
3. Analyzing the new pair < Solact, Sitact > and saving it in the KB for later use.
In order to implement these steps of the CBR method in the area under consideration,
it is necessary to formalize the representation < Sol, Sit >. Meanwhile, it is necessary
to consider that in general, a complex technological object includes elements of
different types, such as technical devices, software and hardware communication and
management systems, servicing and operating organizations (staff), resources, and
other environments.
Further, we will develop a representation of situations by presenting the states of
such elements and the relationship between them.
At the same time, by the situation at a complex object, we will understand the state
of affairs, which is characterized by the current state of the elements of the object and
the connections between them.
3. Results
3.1. The complex technological object structure
The CTO structure contains elements of different types. We will single out the
following elements: equipment, personnel, software and information complex,
resources, buildings, environmental objects and phenomena. Buildings, environmental
objects and phenomena belong to the environment, but are considered as part of the
CTO as they have a connection with the CTO and are able to influence it.
In the ontological representation, a complex object CTO is described by the
quaternion , where O is a set of elements. These elements include:
equipment, personnel, software and information complex, resources, buildings,
environmental objects and phenomena;
S is a set of states:
S={ S ij | ∀i ∈ I ; ∀j ∈ J i}, (1)
where I is a set of indices of CTO elements,
Ji is a set of the ith element state indices. A typical set of states includes "Running",
"Stopped", "Operational", "Not operational", "Present", "Absent", "Available",
"Unavailable," etc.
R is a set of relationships between elements of a complex object:
R={ Rk | ∀k ∈ K}, (2)
where K is a set of indices of relationships between the elements of a CO; it contains
typical relationships Part-of, Has-a, Kind-of, etc. Additional relationships characteristic
of a particular object can be added as well.
A is a set of axioms representing certain necessary combinations of connections
between the elements of the object.
3.2. Representation and retrieve of situations in the case base
Representation of the situation at CTO. In terms of the developed ontological model,
a situation at a complex object will be understood as a set of elements states and a
connection between them at a particular moment. In other words, the Sit z situation is a
projection of an ontological model to a specific environment, which identifies the
specific values of elements, connections, and states:
Sitz=, (3)
where: Oz ⊆ O, Sz ⊆ S, Rz ⊆ R, z is an index of the situation in the set of situations
stored in the knowledge base.
Thus, from the overall representation of a complex object with the GOM, we
transition to presenting situations that reflect the states of the CTO elements and the
relationship between them. Location of elements in (3) according to current data allows
to identify the current situation Sit Act, after which it becomes possible to compare it
with the Sit z situations recorded in the knowledge base.
Retrieve the situational knowledge base. Two similarity metrics are used to retrieve
close situations: similarity in the space of relationships SSim (structural similarity) and
similarity in the space of states PSim (parametric similarity). The first will reach its
maximum value in the event that in two situations there is a complete similarity between
the links of the elements and the minimum value, if in two situations there are no
identical connections between the elements of a complex object. The second will reach
its maximum value when all CTO elements in each of the situations are in the same
state and the minimum value when all the elements of the object are in different states.
First, we will consider the set of relationships on the CTO elements. We will
introduce a graph Gk, which will show kth relationship of the elements of a complex
object. The union of relationship graphs represents the entire set of interaction
relationships in the ontological model.
Further, we will represent the relationship graph Gk with the adjacency matrix M, in
which cells contain 1 if the corresponding elements of the object are in a relationship
from a set Rz and 0 otherwise.
Let M k,act be a matrix for the kth relationship in the current situation, and M k,z be a
matrix for the kth relationship for the zth situation from the knowledge base.
Then we can determine the similarity matrix of the two situations in a relation Rk:
M k (z, act) = M k,z * M k,act , (4)
Where:
* is the operation of elemental multiplication of matrices.
The following formula is used to assess the similarity Simk to Rk:
Simk (Sit Z, Sit Act) = N / max {Nact; Nz}, (5)
Where:
N is the number of non-zero cells in the matrix M k (z, act), this number shows the
number of connections between the elements that are in the first and second situations;
Nact, Nz are the number of non-zero cells in the matrices Mk, act and M k,z, which means
the number of non-zero connections between elements in situations Sit Act, Sit z,
respectively. In this manner, Simk (.) demonstrates the share of matches between two
situations in a relations Rk.
Then the overall estimate of similarity in the space of the relationship is calculated
using a weighted sum:
SSim (Sit Z, Sit Act) = ∑ k Simk(Sit Z, Sit Act), (6)
Where:
k [0, 1], ∑k = 1 is the coefficient of relative importance of kth relationship and
the SSim value lies in the range of 0 to 1.
A similar approach is used to assess similarity in the state space.
A situation matrix representation is also used to assess the similarity of the situation
in the state space. In this case, in the situation matrix, the rows denote the elements of
a complex object, and the columns denote the states in which these elements can be
located. Then, if the element Oi is in the state Sj , then 1 appears at the intersection of
the corresponding row and column.
Line-by-line comparison of two matrices allows calculating the similarity between
the elements Oi in the state space PSimi (SitZ, SitAct). In this case, PSimi (SitZ, SitAct)
takes on the value 1 - if it is in the same position in both matrices, i.e. the elements are
in the same state and 0 otherwise. Then the general affinity between situations in the
state space is:
PSim(Sit Z, Sit Act) = ∑ i PSimi (SitZ, SitAct), (7)
Where:
i [0, 1], ∑i = 1 is the relative importance factor of the ith element in determining
the similarity of situations.
The final similarity of the situation assessment is formed by a pair:
Sim (Sit Z, Sit Act) = , (8)
Sequential retrieve from the knowledge base is done in two stages:
─ retrieve by the criterion: SSim (Sit Z, Sit Act) (1- ) , where is a certain threshold
on which the number of situations retrieved at this stage will depend;
─ retrieve of the closest situation in the KB by the criterion: PSim (SitZ, SitAct)
max.
The experiment. For experimental verification of the described technique, the case is
considered: a malfunction at a complex object of urban infrastructure. The individual
heating station of the building is taken for a complex object. The technological scheme
is an independent two-circuit heating system, where an external coolant through a heat
exchanger transfers thermal energy to the coolant of the home heating system.
The elements of a complex object are formed into groups:
─ technological (pump, heat exchanger, internal pipeline);
─ providing (software, electricity, other equipment);
─ personnel (electrician, plumber, emergency service);
─ environment (substation, neighboring buildings, natural objects, natural
phenomena).
The base of 300 situations with different states was generated using the developed
algorithm in the Visual Basic code language (VBA). The generated situations were
written as state matrices.
Also, the actual situation (Sit act) was generated and wrote in the form of a state
matrix. The Sit act was designed to simulate a pump stopping against a background the
providing equipment some malfunction. Wherein, the pump has a working condition.
A retrieve of similar situations was carried out from the generated base using the
developed algorithm in the VBA, according to the method described above, by the
formula 7. As a result of the selection, the algorithm produced two situations which
having PSim(SitZ, SitAct) = 0.92, i.e. similarity of 92% with the current one. Both
situations have a difference in the state of only one element. The state matrices of actual
and retrieved situations presents in the tables 1-3.
Table 1. T he state matrices of the actual situation (Sit Act).
available
available
workable
interfere
interfere
stopped
in work
faulty
not
not
Sit Act
Input 1
Heat exchanger 1
Pump 1
Other equipment 1
IT 1
Electricity 1
Emergency 1
Plumber 1
Electrician 1
Buildings 1
Nature object 1
Nature event 1
Location 1
Table 2. T he state matrices of the actual situation No1 (Sit z No1).
workable
available
available
interfere
interfere
stopped
in work
faulty
not
not
Sit Z No1
Input 1
Heat exchanger 1
Pump 1
Other equipment 1
IT 1
Electricity 1
Emergency 1
Plumber 1
Electrician 1
Buildings 1
Nature object 1
Nature event 1
Location 1
Table 3. T he state matrices of the actual situation No2 (Sit z No2)
workable
available
available
interfere
interfere
stopped
in work
faulty
not
not
Sit Z No1
Input 1
Heat exchanger 1
Pump 1
Other equipment 1
IT 1
Electricity 1
Emergency 1
Plumber 1
Electrician 1
Buildings 1
Nature object 1
Nature event 1
Location 1
However, during the analysis, there was a semantic difference between these situations.
In the first selected case, the pump remained working against a background the
providing equipment some malfunction. In the second case, the pump also had a broken
condition like providing equipment.
In practice, the solution for each situation will be different, since a working pump
and a broken pump obviously require different actions. Consequently, a collision arises
when selecting a similar situation from the base.
There is introducing the distance between states to exclude such collisions. The
distance between the states of an element is determined by the following formula:
𝑑 𝑖 = ‖𝑆𝑖,𝑎𝑐𝑡 − 𝑆𝑖,𝑧 ‖, (9)
where:
S i, act , S i, z - the state of the ith element in the actual situation and the zth situation from
the base, respectively.
For the case under consideration, the possible states of the pump are ordered on the
interval [0:1] in such a way that S1 is at point 0, S3 is at point 1, and the other states take
values between them:
d
S1 S2 S3
«Broken» «Working»
«Stopped»
Thus, the degree of closeness of situations in the state space is determined by the
formula:
PSim(SitZ, Sit Act) = ∑ 𝑖 ∗ (1 − 𝑑 𝑖 ) , (10)
where:
i –relative importance factor introduce above, which determines by the formula:
𝑟𝑖
𝑖 = , (11)
∑ 𝑟𝑖
where
ri – the significance coefficient of an element in a complex object;
d i – the distance between the states in which the ith element of a complex object is
in the compared situations.
The ri coefficients are formed expertly using the significance scale, in which the
following values are used: 0 - not significant, 1 - weak significance, 3 - medium
significance, 5 - strong significance.
Thus, after the modernization of the formulas due to the distance d, cases were
excluded when two critically different situations with different solutions are issued in
response to the appearance of the same Sit Act.
4. Discussion
In this study, the authors proposed universal models of formalized representation of
situations and solutions for managing a complex object based on the CBR method.
Initially, a general ontological model of a complex urban infrastructure object was
developed, which reflects the elements of the object and the relationships between
them. The situation is considered as a projection of the ontological model onto the
specific states of the CTO elements and the relationships between the elements at a
given time. The decision made in the current situation is represented as a discrete
process of transition from the initial state to the target state. The pairs are
recorded in the knowledge base, where each situation is associated with a solution
recommended for it. The presence of such a knowledge base during situational
management of a complex object allows to quickly find solutions in critical situations
applying cases from the knowledge base.
The proposed way of formalizing CTO allows for a broader view of emerging
situations. The inclusion of elements of CTO environment into its formal representation
allows us to take into account not only the technical aspects of the technological object,
but also the influence of many external factors (the state of the surrounding objects,
organizational systems, climatic conditions, etc.) in the presentation of situations and
decision-making. Such comprehensive assessment of urban infrastructure in the context
of its surroundings allows us to consider the object from the point of view of
environmental safety as well.
Each situation in the knowledge base is associated with a solution that can be
recommended to end users: service personnel, emergency crews, operational dispatch
services of the city and the operating organization. In general, this solution can be
presented as the following set of components:
─ R1, instructions on how to act in a situation (technological map);
─ R2, list of contacts of responsible persons and necessary organizations;
─ R3, required reporting documents (templates, forms, acts, etc.);
─ R4, additional background information (references to similar situations, expert
recommendations, etc.).
It is assumed that during management of complex interactions between different
services, components are addressed to different executors and controllers involved in
resolving situations.
Proposed models allow to organize a view of situations by presenting the final set of
states of the elements and the relationship between the elements of a complex object.
Accordingly, the comparison of the two situations during the retrieve in the knowledge
base will go through the comparison of elementary states and relationships. Of
particular interest is the level of states.
Detailing to the level of elementary states reduces the complex task of comparing
situations to simpler tasks of comparing (recognizing) elementary states. The number
of such states is limited and does not change during the operation of the system.
However, their combination across the many elements of a complex object allows us to
get an almost unlimited in practice set of situations. Various metrics and technologies,
including trained neural networks, can be used to solve these state comparison
(recognition) tasks. This will complement the methods of retrieve situations based on
machine learning metrics, in which the correctness of choice is based on past experience
and training data.
The use of neural networks to implement the CBR method may be one of the areas
of further development of this research. The fact is that distance metrics, which require
calculating multiple parameters [13-15], are traditionally used for comparing situations
in CBR. At the same time, the results can be seriously influenced by both the choice of
the metric and the quality of parameter measurement. The states of different elements
of the object will be described in a variety of ways, from the exact values of numerical
parameters to the quality of the characteristics and graphic images (on maps,
photographs, etc.). For each of these methods, its own state comparison technology can
be used. This opens up possibilities for creating hybrid choice models, where both
metrics and recognizing neural networks will be used to retrieve situations in the
knowledge base.
5. Conclusion
The study produced the following key results: structure of complex object, models of
formalized representation of situations, as well as an approach and metrics for the
retrieve of situations that take into account the structural and parametric similarity of
them were proposed.
These results are important for further development of the methods and technologies
of applied CBR systems. For example, on the basis of these representation models,
inference problems can be solved under conditions of uncertainty of the states of a
complex object and the relations between them. Representation of a complex situation
through a combination of elementary states from a limited set creates the basis for using
neural networks to retrieve cases in the knowledge base of a CBR system through
recognition of its elementary states. Thus, this study is important for the development
of the neurosymbolic artificial intelligence approach as applied to the tasks of managing
complex organizational and technical objects.
In order to further develop the results, we plan to address such tasks as developing
methods for analyzing and comparing states described in different parametric spaces,
as well as generalizing the results for uncertainties of complex object states and the
relations between them.
6. Acknowledgments
The research was funded by RFBR and Tyumen Region, project number 20-47-
720004.
References
1. Kiziroglou, M. E., Boyle, D. E., Wright, S. W., Yeatman, E. M. Acoustic power delivery to
pipeline monitoring wireless sensors. Ultrasonics, 77, 54 -60 (2017).
2. Zarifi, M. H., Deif, S., Daneshmand, M.: Wireless passive RFID sensor for pipeline integrity
monitoring. Sensors and Actuators A: Physical, 261, 24-29 (2017).
3. Junie, P, Dinu O., Eremia C., Stefanoiu D., Savulescu I.: A WSN Based Monitoring System
for Oil and Gas T ransportation through Pipelines. IFAC Proceedings Volumes 6(45), 1796-
1801 (2012).
4. Zrelli, A., Ezzedine, T .: Design of optical and wireless sensors for underground mining
monitoring system. Optik, 170, 376-383 (2018).
5. Zengqiang Xing at al.: Design and Application of T ransmission Line Intelligent Monitoring
System. E3S Web of Conferences 185, 01063 (2020).
6. Jia, Z.: Pipeline abnormal classification based on support vector machine using FBG hoop
strain sensor. Opti. 170, 328-338 (2018).
7. He, P., Wang, J.: Statistical process monitoring as a big data analytics tool for smart
manufacturing. Journal of Process Control 67, 35-43 (2018).
8. Ariyaluran Habeeb, R. A., Nasaruddin, F., Gani, A., Hashem, I., Ahmed, E., Imran, M.:
Real-time big data processing for anomaly detection: A Survey. International Journal of
Information Management, 45, (2018).
9. Rukmani, P., T eja, G.K., Vinay S., Reddy B.P.K.: Industrial Monitoring Using Image
Processing, IoT and Analyzing the Sensor Values Using Big Data. Procedia Computer
Science, 133, 991-997 (2018).
10. Z. Sh. Juraev, D. T . Muhamediyeva, D. M. Sotvoldiev: Construction of hybrid intellectual
monitoring and decision-making systems. Journal of Physics: Conference Series, 1546,
012083 (2020).
11. Andrzej Bieleckia, Mateusz Wójcik: Hybrid AI system based on ART neural network and
Mixture of Gaussians modules with application to intelligent monitoring of the wind turbine.
Applied Soft Computing Journal, 108, 107400 (2021).
12. Andrey Yu. Puchkov, Maksim I. Dli, Ekaterina I. Lobaneva: Application of deep neural
networks in models of complex technological objects. Bulletin of the Saint Petersburg State
Institute of T echnology (T echnical University), 52, 104 -110 (2020).
13. Eremeev, A., Varshavskiy, P., Alekhin, R.: Case-Based Reasoning Module for Intelligent
Decision Support Systems. Proceedings of the First International Scientific Conference
“Intelligent Information Technologies for Industry” (IIT I’16), Vol. 1. Part III, pp. 207-216,
Springer International Publishing (2016).
14. Galina S. Maltugueva, Alexander Yu. Yurin: Improving case-based reasoning with the aid
of multi-criteria and group decision-making methods. Proceedings of the International
Convention MIPRO, IEEE, 1031 – 1036 (2019).
15. Xiaoyan Jiang at al.: A Decision Method for Construction Safety Risk Management Based
on Ontology and Improved CBR: Example of a Subway Project. International Journal of
Environmental Research and Public Health. 17, 3928 (2020).
16. Hoda Nikpour, Agnar Aamodt: Inference and reasoning in a Bayesian knowledge-intensive
CBR system. Progress in Artificial Intelligence. Progress in Artificial Intelligence
volume 10, pages 49–63 (2021).
17. Kolodner, J.: Case-Based Reasoning. Morgan Kaufmann Publishers, Inc., San Mateo
(1993).
18. Eisenstadt V., Langenhan C., Althof K.: Generation of Floor Plan Variations with
Convolutional Neural Networks and Case-based Reasoning An approach for transformative
adaptation of room configurations within a framework for support of early con ceptual
design phases, Proceedings of 37 eCAADe and XXIII SIGraDi Joint Conference, p. 79-84
(2019).
19. Mathisen, B.M., Bach, K. & Aamodt, A.: Using extended siamese networks to provide
decision support in aquaculture operations. Applied Intelligence (2021).
20. M.F. Abdelwahed et al. : Solving the motion planning problem using learning experience
through case-based reasoning and machine learning algorithms. Ain Shams Engineering
Journal, 11, 133–142 (2020).