CC BY
25Principles of Constructing Artificial Intelligence Systems and their Application in Electrical Power Industry
Alexander Yu. Khrennikov1*, Yuri Ya. Lyubarsky2, Andrei Yu. Khrennikov2
1 JSC "S&T Centre of Federal Grid Company of Unified Energy System" Rosseti, Moscow, Russia
2 Mathematical Institute of Linnaeus University, Vaxjo, Sweden
Abstract — The paper considers two historically established directions of building artificial intelligence systems: expert systems and neural networks. The consideration is given to parallels and analogies between the work of neural networks and the work of the human brain, the hierarchical structure of the brain, the history of the neurocybernetics development, and the theoretical foundations of neural networks.
Models of reasoning of specialists are used in expert systems based on production rules (expressions of the form "if ..., then"). The main semantic elements in neural networks are models of neurons and layers of neurons (input, output, intermediate). Large amounts of data are associated with neuron models (the so-called "big data").
The paper gives examples of using neural networks and artificial intelligence systems in the electric power industry for the electrical equipment state monitoring, operational dispatch control of electrical networks, for the analysis of emergencies, and intelligent functions of automated dispatch control systems (ADCS).
Index Terms: artificial intelligence systems, expert systems, neural networks, hierarchical structure of the brain, neurocybernetics, electrical equipment state monitoring, operational dispatch control of power grids, emergencies, automated dispatch control systems (ADCS).
* Corresponding author. E-mail: [email protected]
http://dx.doi.org/10.38028/esr.2021.04.0006
Received November 18, 2021. Revised November 29, 2021.
Accepted December 01, 2021. Available online January 25, 2021.
This is an open access article under a Creative Commons Attribution-NonCommercial 4.0 International License.
© 2021 ESI SB RAS and authors. All rights reserved.
I. Introduction. Problem statement
Once, many years ago, Descartes, looking through a barred window at a growing oak in the courtyard, realized that with the help of a window lattice it was possible to specify the positions of parts of an oak (trunk, branches, leaves) by numbers, i.e., to digitize an oak! By reducing the mesh size of the lattice, which will have more and more details, one can digitize an oak. Descartes exclaimed: "Eureka!" and created a rectangular cartesian coordinate system. This was the moment of paramount importance in the mathematization of physics and the beginning of digitalization. Any material object could be encoded using Cartesian coordinates. The motion of this object could be described by functional transformations of Cartesian coordinates. We can say that a numerical image of physical space was created. Today's digitalization began with that very event.
This paper discusses two historically established directions of building artificial intelligence systems [1-3]:
expert systems, neural networks.
Neural networks and expert systems are a large class of systems, and their architecture is analogous to the construction of neural tissue from neurons. One of the most common architectures, a multilayer perceptron with error backpropagation, simulates the operation of neurons as part of a hierarchical network, where each higher-level neuron is connected by its inputs to the outputs of neurons of the underlying layer [1].
Logical and symbolic operational disciplines have dominated artificial neural networks in recent years. For example, expert systems have been widely promoted and have notable success, as well as failures. Some scientists state that artificial neural networks will replace modern artificial intelligence, but there is much evidence that they will combine into systems, where each approach is used to solve the problems it copes with better [2].
This view is reinforced by the way people function in the world. Pattern recognition is responsible for the activity that requires a quick response. Since actions are performed quickly and unconsciously, this mode of functioning is critical for survival in a hostile environment. Imagine just what would happen if our ancestors were forced to consider their reaction to a jumping predator?
When our pattern recognition system is unable to give an adequate interpretation, the question is transmitted to the higher parts of the brain. They may ask for more information and take longer, but the quality of the resulting solutions may be higher.
The values of the input parameters are fed to the neurons of the lowest layer, which is why it is necessary to make some decisions, predict the development of the situation, etc. These values are considered signals transmitted to the next layer, weakening or amplifying depending on the numerical values (weights) attributed to interneuronal connections. As a result, some value is generated at the neuron output of the uppermost layer, which is considered a response, i.e., reaction of the entire network to the entered values of the input parameters.
For the network to be used in the future, it must first be "trained" on the data obtained earlier, for which both values of input parameters and correct answers to them are known. Training involves selecting the weights of interneuronal connections that ensure the closest proximity of the network responses to the known correct answers [1-5].
The object of the scientific research of this paper is to review and investigate the development of artificial intelligence (AI) systems in the electric power industry. As noted in the Analytical Review prepared by the Russian Energy Agency of the Ministry of Energy of Russia in terms of the frequency of references in various publications around the world, the functional AI applications are divided into nine main groups:
1. Methods of management (growth + 55% over the last 4 years);
2. "Machine vision" (growth by 49%);
3. "Distributed intelligence;"
4. Natural Language Processing (+ 33% growth);
5. Representation of knowledge and logical judgments;
6. Planned behavior (growth + 37%);
7. Predictive (predictive) analytics;
8. Robotization (growth + 55%);
9. Processing of speech information (+ 15% growth). The most active use of artificial intelligence
applications was recorded in the following groups:
• Transport (15% of all patents, + 30% growth);
• Telecommunications (15% of all patents, + 24% growth);
• Biology and medical research (12% of all patents). Agriculture (+ 30%) and settlement of government
tasks (+ 30%) are also rapidly growing application areas.
Energy tasks are separated into a separate area of use (energy management).
Examples of the artificial intelligence methods used in energy are listed in [6]. The most promising task groups, where AI can have an effect, are:
• forecasting tasks (meteorological information, equipment operation status, energy consumption changes, etc.);
• optimization tasks (operating conditions of power system components, consumption, network configuration, etc.);
• management tasks (artificial lighting, renewable energy sources, batteries, asset performance, etc.);
• communication tasks (energy companies with consumers);
• the tasks of developing services and services (in terms of customer satisfaction with the range of services provided by companies, participation of enterprises in the energy markets, solving issues of quality assurance).
• It is noted that the expansion of the use of artificial intelligence tools in the energy sector will inevitably occur along with such processes as:
• energy transformation due to the expansion of the use of local renewable energy sources, as well as the smartization of energy generation,
• transmission, and consumption (smart technology);
• digital transformation due to the growing needs of monitoring and data analysis ("big data") and the introduction of new technologies (for example, blockchain, "digital substation," unmanned devices for monitoring objects, etc.);
• unification and mutual influence of various energy sectors and transport sectors (for example, Power-to-X technology) [6].
Thus, the main focus of this paper is on the principles of building artificial intelligence systems and their application in the electric power industry.
This paper has the following structure: introduction; parallels and analogies between the work of neural networks and the work of the human brain; the hierarchical structure of the brain and modeling of thinking processes in p-adic coordinate systems; general principles of constructing artificial intelligence systems and their application in the electric power industry.
II. Parallels and analogies with the work of the human brain
Neural networks are designed similar to the human nervous system, but they use statistical analysis to recognize patterns from a large amount of information through adaptive learning. The nervous system and the human brain are composed of neurons connected by nerve fibers. Nerve fibers are capable of transmitting electrical impulses between neurons. Processes of transmitting irritations from our skin, ears, and eyes to the brain; processes of thinking and controlling actions - all of them are implemented in a
neural axon to reach the synapse. Neurotransmitters regulate the passage of electrical signals through synapses.
Fig. 2. A functional MRI image showing areas of high mental activity. The lower image shows a flower-like structure created by a diffusion MRI machine tracking neural pathways and brain connections.
living organism as transfer of electrical impulses between neurons.
Each neuron has two types of nerve fibers - dendrites, along which impulses are received, and a single axon, along which the neuron can transmit impulses. Axon contacts the dendrites of other neurons through special structures, synapses, which affect the impulse strength (Fig. 1) [3-5].
Artificial intelligence systems and expert systems are too "fragile," as these systems encounter a situation not envisaged by the developer, they either form error messages or give incorrect results. In other words, these programs can be quite easily "confused." They cannot continuously self-learn, as a human does while solving emerging problems.
In the mid-1980s, many researchers recommended using neural networks to overcome these (and other) disadvantages.
A neural network, in its most simplified form, can be considered as a way of modeling the principles of organization and functioning mechanisms of the human brain in technical systems. Modern concepts suggest that the human cerebral cortex is a set of elementary interconnected cells (neurons), the number of which is estimated at 1010. Technical systems, in which an attempt is made to reproduce such a structure (hardware or software), albeit on a limited scale, are called neural networks.
A neuron in the brain receives input signals from many other neurons, and the signals have a form of electrical impulses. Neuron inputs are divided into two categories - excitatory and inhibitory. The signal received at the excitatory input increases the neuron excitability, which, when a certain threshold is reached, leads to the formation of an impulse at the output. The signal arriving at the inhibitory input, on the contrary, reduces the excitability of the neuron. Each neuron is characterized by an internal state and a threshold of excitability. If the sum of signals at the excitatory and inhibitory inputs of the neuron exceeds this threshold, the neuron generates an output signal that goes to inputs of other neurons connected to it, i.e., there is a propagation of excitation through the neural network. There are at least 30 billion neurons in the brain, and each of them can have up to 10 thousand connections with other neurons.
The switching time of an individual neuron in the brain is of the order of a few milliseconds, i.e., the switching process is slow. Therefore, the researchers concluded that the high performance of information processing in the human brain can only be explained by the parallel operation of many relatively slow neurons and a large number of mutual connections between them. This explains the widespread use of the term "massive parallelism" in the literature on neural networks [1-5].
To create a model of the work of the human brain, then to move on to the creation of an artificial intelligence system, one can try to use the same coordinate system to describe the mental world that was used to describe the
material world. The overwhelming majority of researchers follow this path. Hundreds of laboratories worldwide use more and more sophisticated magnetic resonance devices to make Cartesian card activation of neurons in the brain increasingly more accurate. This is an interesting activity. There are doubts, however, that it can lead to understanding the mental processes and the work of the brain. Recently, numerous studies have also been conducted to create quantum models of the brain. This direction is right. Making sure that the processes occur in the human brain, it is impossible to put it in a system of coordinates of the Euclidean space R3, it is necessary to try to put it in other multi-dimensional spaces, for example, Hilbert space H. However, the vast majority of quantum mental studies are based on an extremely questionable reduction postulate: the mental processes in the brain can be reduced to quantum physical processes in the microworld (Fig. 2) [7-9].
The quantum behavior of the brain is a consequence of its specific informational structure and not the effect of the quantum behavior of microscopic components of the human brain.
Therefore, on the way to artificial intelligence systems, we must create a mathematical model of the mental space, and choose the appropriate coordinate system.
The most important feature of the real continuum is its homogeneity. All points of physical space (with a three-dimensional model R3) are equal. We cannot say that one point is more important than another. However, brain activity is inhomogeneous. We cannot say that all human thoughts, ideas, concepts, feelings are equal. Moreover, the brain is hierarchical. There is a clearly expressed hierarchy of concepts, images, and feelings. However, there is no complete orderliness in the mental space. It is impossible to hierarchically arrange all concepts, images, and feelings. There are non-comparable ("incommensurable") spiritual objects, thoughts, and images [7].
III. History of the issue
The theoretical foundations of neural networks were laid back in the 1940s by W. McCulloch and W. Pitts. In 1943, their work "The Logical Calculus of Ideas Relating to Nervous Activity" was published, in which authors built a model of a neuron and formulated the principles of constructing artificial neural networks.
A substantial impetus to the development of neurocybernetics was given by the American neurophysiologist Frank Rosenblatt, who in 1962 proposed his neural network model - the perceptron. Initially, it was received with great enthusiasm, but soon it came under intense attack from major scientific authorities. And although a detailed analysis of their arguments shows that they did not contest exactly the perceptron proposed by Rosenblatt, large research on neural networks has been curtailed for almost ten years.
Despite this, the 70 years saw a lot of interesting developments, such as cognition capable of recognizing
fairly complex images well regardless of rotation and zoom.
In 1982, the American biophysicist J. Hopfield proposed an original neural network model named after him. Many efficient algorithms were found (a counter-flow network, bidirectional associative memory, and others) in the next few years.
In 1986, J. Hinton and his colleagues published an article describing a neural network model and an algorithm for its training, which gave a new impetus to research on artificial neural networks.
The next book, "Modeling thinking processes in p-adic coordinate systems," deals with the mathematical modeling of thinking processes based on dynamic systems on p-adic trees and more general ultrametric mental spaces. Applications in psychology (including Freud's psychoanalysis) and cognitive sciences are considered in [7].
IV. Hierarchical structure of the brain
In modern neurophysiological and cognitive literature, one can gradually become stuck in R3-map scintillation excited neurons, neural networks work, electricity flows in the brain, i.e., be automatically involved in using mentality's real processes of the Cartesian coordinate system (developed to investigate Matter) for research. However, Sigmund Freud did not write about the functioning of neural systems. He described streams of ideas, representations, and desires, and these mental objects in the Freudian description are no less real than material objects. Mental objects evolve, interact with each other; mental forces are active here. For example, one of these forces displaces strong (often shock) but forbidden experiences into the unconscious, thereby creating complexes. The complexes, in turn, induce forces acting from the subconscious on the streams of conscious ideas.
Intuitively, it is clear that we are dealing with dynamics in mental space, very similar to the dynamics of material objects in physical space. It is only necessary to introduce an appropriate system of mental coordinates and describe the mental flows mathematically. For reasons that have already been given above, there is no capability to use models. As noted, mental space is not homogeneous; it is also not ordered: we cannot compare two arbitrary mental objects. On the other hand, there is a clear hierarchical structure in the mental world. Note that disorder is quite consistent with hierarchy. For two mental objects x and y, there always exists some mental object z, which stands in the hierarchical system above x and y. However, in this case, x and y can be incomparable with each other [7].
We get a (in principle, infinite) tower of mental spaces. This is a new type of cerebral mental hierarchy, i.e., the hierarchy of mental spaces. Recall that each of these spaces is a hierarchical tree. In particular, if the original tree is identified with tree /-states generated by the brain, we obtain an infinite hierarchical tower (hierarchical) mental
space, which "relies on" the brain. Of course, we currently do not have experimental data (from neurophysiology and cognitive sciences) that could be interpreted as evidence of the existence of a vertical hierarchy of mental spaces. Some indirect evidence of the existence of towers of mental space, or rather, the brain's ability to operate simultaneously in several mental spaces, i.e., to process cognitive information of different mental levels in parallel.
Note that the parallelism introduction in this model of a hierarchical tower of mental spaces does not mean independence. The /-states of the lowest level form associations, which are the /-states of the next level. Thus, the most primitive /-states enter through the hierarchy of mental spaces into /-states of the highest mental level. It is possible that functioning in a tower of mental spaces is the basis of the brain's truly limitless informational capabilities.
This model does not exclude that, for example, the human brain can function in an endless tower of mental spaces. This will mean that the final physical system - the brain - can have endless information power. But even if the brain can use only a finite number of K floors of the hierarchical tower of mental spaces, then its information power is significantly greater than the power of a cognitive system using only the first floor, for example, only /-states produced by a neural network. Calculations show that information power grows linearly with increasing K. The value of K can be used as a numerical characteristic of the level of mental development. It is quite natural to assume that K grew in the course of evolution, reaching the highest of the %value in animals and humans. Moreover, K may depend on the species or even the individual. Although, in principle, the emergence of the human mind could correspond to the jump from finite K to infinite [7].
Leaving aside the hypothetical possibility of creating hierarchical mental towers, starting from some fixed ultrametric mental space X, let us return to a model that uses only X. It is natural to assume that different cognitive systems generate various spaces X. In particular, mental spaces can be generated in the form of different p-adic trees. There are 2-adic, 3-adic, and ... cognitive systems. However, it does not follow from the above that, for example, each person has their p (2-adic person, 3-adic person, ...). Different subsystems of the same brain can generate different p-adic trees. For example, there can be mental spaces arising in the form of X = Q5 x Q7 x Qn, as well as more general ultrametric spaces.
After the Russian Television broadcast "At Gordon's," where the professor of mathematics expounded p-adic models of thinking in a conversation with one of the founders ofp-adic physics, also a professor of mathematics, a security guard at Steklov Institute of Mathematics asked his interlocutor: "Tell me, professor, but are my brains 2-adic or, for example, 7-adic?"
One can use the general theory of metric spaces to describe mental processes. In particular, the possibility
of representing physical metric spaces using mental ultrametric spaces is under study. The inverse problem of embedding an ultrametric mental space into a physical Euclidean space is also considered. An amazing topological fact (theorem of A. Lemin) is the impossibility of embedding an ultrametric space containing n + 1 points into Rk, k < n. In particular, only a mental space containing four points can be embedded into the Euclidean space R3. Even a five-point mental space cannot be nested in R3.
Thus, the physical representation of a mental space containing hundreds of thousands of mental points requires a Euclidean space of unthinkable dimensions. In particular, the physical Euclidean image of the human mental space - the brain - arises from projecting a huge number of mental points onto each point of the physical area of the brain. Note that an infinite p-adic tree can be isometrically embedded only in an infinite-dimensional Hilbert space.
In what direction to proceed further on the path to creating an artificial intelligence system? It is crucial to simulate the higher mental activity: model selection, training, manic and depressive states, homeostatic states, habituation, imprinting, i.e., instant subcortical learning.
The most important feature of the Freudian description of the mental world is the reality of this world. For Freud, ideas, desires, feelings, experiences are no less real than, for example, mountains, houses, horses, ... For Freud, mental processes are no less real than physical and chemical processes. Ideas, desires, feelings, emotions, and experiences interact in the same way as physical bodies do. For Freud, mental forces are no less real than physical forces: "I have therefore confirmed that forgotten memories have not disappeared. The patient is still possessed by these memories and they are willing to enter into an association with what he knows, but some force prevented them from being conscious and forced them to remain unconscious. The existence of such a force could be accepted quite confidently since the corresponding strain was felt when trying, in contrast to it, to bring unconscious memories into the patient's consciousness. One felt the strength that supported the painful state, namely, the patient's resistance" [8].
By introducing the corresponding field of forces, we get the corresponding mental field. Thus, it is quite natural to try to encode all mental objects that arise in Freud's psychoanalytic scenarios with the help of some mathematical structures and try to model the mental processes, in particular, the emergence of complexes within the framework of mathematical models. Undoubtedly, this is a highly complex problem and the p-adic model can describe only some features of mental behavior.
The most important postulate of Freudianism is the one of the determinism of mental processes. Roughly speaking, the mental trajectory is determined by the initial conditions (for example, childhood experiences) and the corresponding field of mental forces.
The mental field can be treated as an information field. The natural question about its measuring arises. Another question is about coupling with physical fields, e.g., the electromagnetic field generated in the brain. An important step in this direction was recently done in [9]. Here, EEG signals from the brain were transformed with the aid of a clustering algorithm into dendrograms, which, in turn, can be algebraically represented by p-adic numbers. A corresponding "mental field" was reconstructed as Bohm potential on the p-adic configuration space. This approach was used for medical diagnostics of epilepsy.
No diagnostic or predictive instruments to help with early diagnosis and timely therapeutic intervention are available as yet for most neuropsychiatric disorders. A quantum potential mean and variability score (QPMVS) was developed to identify neuropsychiatry and neurocognitive disorders with high accuracy, based on routine EEG recordings. Information processing in the brain is assumed to involve the integration of neuronal activity in various areas of the brain. Thus, the presumed quantum-like structure allows quantification of connectivity as a function of space and time (locality) as well as instantaneous quantum-like effects in information space (non-locality) [9].
V Making the final decision in the brain. How to CREATE effective ARTIFICIAL INTELLIGENCE
In [10], Michio Kaku compares the brain to a large corporation. This idea has a right to exist; it can explain some interesting properties of the brain:
• The bulk ofthe information is in the "subconscious," i.e., the general director, fortunately, has no idea about the deep streams of information continuously circulating through bureaucratic channels. Moreover, only a tiny fraction of the information ultimately ends up on the desk of a senior executive, who can be compared to the prefrontal cortex. The general director only gets to know the data that is important enough to merit his attention; otherwise, his activity would be paralyzed by the avalanche of unnecessary information.
• Probably, such an organization of brainwork is a byproduct of evolution since, under critical conditions, our ancestors could not afford to overload the brain with superficial subconscious information. Fortunately, we do not notice all the trillions of operations that our brain constantly performs. Having met a tiger in the forest, one does not have to think about the state of their stomach, toes, hair, and others at that moment but just needs to remember how to run faster.
• "Emotions" are quick decisions that are self-born at a low level. Since rational thoughts take a long time, and, in a critical situation, there is no time to think, low-level areas of the brain must quickly assess the situation and make a decision (generate emotion) without permission from above.
Thus, emotions (fear, anger, horror, etc.) are instantly
appearing alarm flags, the command to which is given at a low level and the purpose of which is to warn the control center about a potentially dangerous or difficult situation. Consciousness has virtually no control over emotions. For example, no matter how we prepare for a public speech, the nervous tension will not disappear [10].
Rita Carter, the author of "Mapping the Mind" paper, writes: "Emotions are not feelings at all, but a set of physiological survival mechanisms that emerged as a result of evolution. Their task is to direct us away from danger, towards what may be useful" [11].
There is a constant struggle for the attention of the leader, but there is no single homunculus, central control panel, or Pentium processor making decisions; instead, various local centers within the leadership are in constant competition with each other for the director's attention. Therefore, thoughts do not go in a smooth continuous sequence, all kinds of feedbacks compete with each other, thereby creating a real cacophony. The concept of "I" as a single integral entity, continuously making all decisions, is just an illusion generated by the subconscious.
We ourselves feel that our consciousness is one, that it continuously and evenly processes information and fully controls all our decisions. However, brain scans give a completely different and objective picture.
Professor at the Massachusetts Institute of Technology (MTI) Marvin Minsky, one of the founding fathers of the Laboratory for Artificial Intelligence, said that the human mind is more like a "society of minds," consisting of submodules that are constantly fighting among themselves.
Harvard psychologist Steven Pinker explains how consciousness arises from all this confusion: consciousness is like a storm raging in my head. "John intuitively feeling that there is a guideline of "I" that sits in our brain's control center, looks at the screen with the data from the senses, and pushes buttons to send commands to our muscles is just an illusion. Rather, consciousness is a vortex of events distributed throughout the brain. These events compete for attention, and when one of them manages to out-shout all the others, the brain rationally substantiates the result retroactively and fabricates the impression that everything happened under the control of a single center."
Final decisions are made by the general director in the control center. Almost all the bureaucracy exists to collect and organize information for the general director, who only meets with the department heads. He tries to bring to a common denominator all the conflicting information coming to the command center. All intrigue ends here. The general director in the prefrontal cortex has to make the final decision. Whereas in animals most decisions are made instinctively, humans make high-level decisions after carefully analyzing and sifting through the information coming from the senses [10].
Information flows are hierarchical. Since a huge amount of information must pass both up to the management office and down to the performers, this information must be
_user interface
* —
output subsystem
r
knowledge base
_!_
data «
knowledge representation model <
expert
_!_
knowledge engineer
r
intelligent editor
Fig. 3. Structure of the expert system.
organized into complex systems of embedded networks with many branches. Imagine a spruce tree with a guiding center at the top and a pyramid of branches below, leading to many less significant centers.
There is, of course, a difference between the bureaucratic system and the structure of thinking. It is known that the first rule of any bureaucracy is that "it expands and fills all the allocated space". Wasting energy, however, is a luxury that the brain cannot afford. The brain only consumes about 20 watts of power (like a weak incandescent lamp), but this is probably the most it can take without depriving the rest of the body of functionality. If the body generates more heat, the tissues will fail. Therefore, the brain constantly conserves energy and uses all sorts of tricks for this. These clever ways are invented by evolution to simplify various actions.
VI. Artificial intelligence systems
Consider two historically established directions of building artificial intelligence systems [1-3]:
• expert systems,
• neural networks.
The historical aspects of the development of these directions will not be presented here (this was done approximately in the same period but by completely different groups of specialists). The distinction between directions is somewhat confused by the partial commonality of terminology (such as "artificial intelligence," "decision making," and others). Genuine differences are revealed
Fig. 4. The architecture of the expert system.
Fig. 5. An example of a simple neural network.
Fig. 6. Structural diagram of a remote control devicefor leakage current of the surge arrester.
if we pay attention to the semantics of the main class of concepts in the corresponding intelligent models (Fig. 3).
In expert systems [12-14], these are models of reasoning for specialist people. Reasoning models are often based on production rules (expressions like "if... then"). Access to the rules on the limited natural language allows accompanying the decision-making in expert systems with explanations that are clear to users. This method is used to create some intelligent expert systems, including energy applications. An example of the architecture of the expert system is shown in Fig. 4.
In neural networks, the basic semantic elements are models' neurons and neuron layers (input, output, intermediate) (Fig. 5) [12]. Large amounts of data are associated with neuron models (the so-called "big data"). A partially trained neural network can be used for other tasks as well. The successful application of neural networks is known in strategic games, recognition of images, etc. A significant drawback of the neural network application is a complete absence of explanation of the system's actions. Knowing how apply layers of neurons to new tasks is a skill close to the invention. A typical example is various variants of human face recognition [12-14].
Speaking about the speed of intelligent systems, one should note the fundamental advantage of expert systems - they can use "operational results" to start a solving process, while in neural networks, one has to start from afar (from the neuron model).
Currently, most experts believe (based on the achievements of neural networks in games and in the military field) that neural networks have finally won the competition in intellectual areas. A more cautious conclusion, however, seems attractive: for "routine" operational tasks (such as automated control systems for electric power grids), neural networks should be considered priority expert systems (tasks of analyzing graphic images should be attempted by other methods). There is no alternative to neural networks when working with "big data" [5-14].
Let us consider several examples of using neural networks in the electric power industry to monitor the state of electrical equipment.
VII. Remote control device for leakage current of
SURGE ARRESTER
The current flowing through the surge arrester (leakage current) is a geometric sum of capacitive and active components, with the predominant component of the capacitive current during normal operating conditions (Fig. 6).
This device has several advantages, unlike standard diagnostic methods for surge arresters (conduction current measurement):
1. Wireless communication of measuring sensors and measuring module (up to 1000 meters).
2. Efficiency in the processing and transmission of data received in measuring sensors to the user.
3. Short response time to changes in measured values and quick notification of the user in case of abnormal changes in measured data.
4. Sensors form a system for receiving and transmitting data with the help of transmitting and receiving modules.
5. Stability of the system. Failure of a sensor (several sensors) does not lead to a malfunction of the system.
6. The number of measuring sensors working with one measuring module (personal computer) can be 100 or more.
7. The total length of the system can be tens of kilometers.
8. Ability to connect to an Ethernet network via TCP/IP protocol to obtain remote access to the system at the request of the customer.
9. The GPS module, a global positioning system, when used in the sensors, provides information about the exact geographic location of the sensors and real time, which allows them to synchronize their work [15]. The current passing through the surge arrester flows
through the grounding conductor, in the cut of which the
f
Fig. 7. An example of organizing the communication route of the current .sensor No. 14 with the basic module.
measuring module (MM) is installed. Current is measured using a resistive shunt located inside the MM. Data from the MM is transmitted using a transceiver device based on ZigBee technology. The MM conducts primary current measurement, temperature measurement and determines the triggering of surge arrester when the current exceeds a certain value. The basic module (BM) communicates with the measuring modules and configures the MM network. The basic module provides receiving and transmitting functions to transfer information from the MM to a personal computer and data processing in software installed on a PC. The software performs digital signal processing. Maximum effective values of currents (capacitive and active components) are determined.
The time interval of data collection is determined once a month, as well as in case of an abnormal increase in current. The report indicates the number of the current sensor, time, temperature, data array of the values of the flowing current. When the current exceeds a certain value, the time interval is reduced to 2.5 hours. In turn, the operator can request sensor readings at any time.
The wireless data transmission system based on the ZigBee technology allows organizing a radio network by connecting all devices in a single multilocular network. The advantage of the ZigBee technology is its stability. It is capable of organizing a radio network in the event of failure of one or several sensors in a single multimesh
Fig. 8. Topological model of the primary network for dispatch control tasks.
network, and signaling the failure of sensors (Fig. 7) [15].
The authors of [16] present the experience in the development and implementation of decision support systems based on machine learning algorithms in various tasks of the electric power industry, analyze the main errors and the consequences of their influence on the results of the operation of such systems in the electric power industry.
This is done on the example of the problem of predicting the photovoltaic generation introduced into the technological activities of electric power network [16].
VIII. "Smart" electrical grids for dispatching
DECISIONS
For "smart" electrical grids, we considered software tools based on artificial intelligence methods that perform new functions and increase the level of computer support for dispatching decisions [13].
One of the smart grid goals, in this case, is to ensure recovery after a breakdown. Therefore, particular attention is focused on the problems of diagnostics of emergencies, intelligent monitoring of the states of electrical networks, and planning of post-emergency restoration of power supply. A new type of software simulator (a simulator for analyzing emergencies) for dispatchers of electrical networks is considered in [13]
A multi-agent structure of an intelligent automated dispatch control structure was employed as part of artificial intelligence used.
Fig. 9. Logical model of relay protection for dispatch analysis tasks.
The problem of analyzing technological breakdowns and accidents and the possibility of forming (based on the data of the operational information complex) an operational reference about an accident in the electric power system to facilitate and accelerate the investigation of technological breakdowns and accidents at substations (with examples of investigation acts) are also considered.
The concept of "volumetric" decision-making is introduced, reflecting the participation in decision-making of groups of specialists with different competencies. In this case, the concept of an expert system with a "bulletin board" is used.
The possibility of using the concept of extreme programming is considered to facilitate the transformation of the operational experience of technologists into the formalisms of a natural language expert system.
For definiteness, this paper assumes the use of the MIMIR expert system (shell) since this system has several successful applications in electric power problems [12].
"Intelligent" dispatching systems (IDS) [12, 13] are usually called systems that contain, in addition to traditional functions (collection of operational information, maintaining real-time databases and archives, performing calculations, graphical presentation of information in the form of mnemonic diagrams, graphs, diagrams, generation reports) and intelligent functions (based on knowledge [13]), such as:
• situational analysis of the control object, including the analysis of events and situations,
• determination of the necessary actions of the operator in the event of emergencies,
• blocking against unauthorized actions of the operator,
• maintaining knowledge bases of real time [17-20].
Traditional dispatch systems do not provide operating
personnel with a sufficient level of information support for the operational dispatch control of electric power grids in emergencies. Receiving images of circuits with the positions of switching devices and values of electrical parameters indicated on them, the dispatcher must "think themselves" whether the situation is abnormal, what is the cause of the situation, what actions need to be taken to restore the normal situation. The cost of human error in making such decisions can be very high.
The dispatching systems existing for electric power grids (at least domestic ones) have no intelligent functions. At present, the prerequisites for changing this situation have emerged, there are real-time domestic expert systems tested in operation [17, 18]. Ensuring that information about relay protection and control systems is entered into the IDS is no longer a serious problem (see, for example, Siemens ACS - proavtomatika.ru).
The implementation of the IDS functions is achieved by including some intelligent agents (IA) in the system built programmatically based on the technology of expert systems, and algorithmically - on a set of technological instructions.
Such a system with elements of artificial intelligence should recognize precisely abnormal situations, separating them from standard ones, such as, for example, shutting down equipment elements to be taken out for repair on request.
When analyzing an abnormal situation (technological
breakdowns and accidents at substations), the system must determine:
• source of technological disturbances (for example, short-circuit on one of the equipment elements);
• work of automatically repeated switching-on (successful or unsuccessful);
• failures in the operation of circuit-breakers;
• tightening of the switch of circuit-breakers;
• failures in the operation of relay protection and automatic controls;
• excessive or non-selective operation of relay protection and automatic control devices;
• disconnection of equipment elements (transmission lines, transformers, bus-bars).
This data is minimally necessary for dispatching and operating personnel to assess the situation in the electrical network and begin planning measures to prevent the development and eliminate technological disruptions.
Some scientific papers [17-26] consider the intelligent functions of automated dispatch control systems (ADCS). The effectiveness of these functions was shown when describing the situations and generating advice for dispatching personnel in emergencies. These functions use the platform of expert systems, and the formalism of forming the reasoning models of technologists is applied [16]. The real-time operation of the ADCS complex requires that the architecture of the software of this complex be considered. At the same time, relying on a variety of functionally different workstations, it is advisable to distribute intelligent functions so that they correspond to the experience of various technologists and, at the same time, only the part of the overall task "adjacent" to the experience of the relevant specialists (in particular, the analysis of the situation) is performed. This approach is consistent with the concept of multi-agent systems of artificial intelligence [12, 20-26].
Intelligent programs for analyzing the situation for ADCS are functionally different and use different sections of the system's knowledge. It is appropriate to organize the execution of these programs by different intelligent agents, i.e., relatively simple modules, each of which uses its own section of knowledge and has its own conditions of initiation. Since ADCS is normally implemented as a computer network, intelligent agents are localized in computerized workstations for various specialists.
Figure 8 shows an example of a semantic network structure for dispatching tasks in electric power grids. Concepts of "semantic network to represent relay protection" are combined into homogeneous semantic groups. Interaction between users and an intelligent system can be organized with the help of a limited natural language.
This interaction is based on natural language issues. A semantic network for representing relay protection and automatic controls is shown in Fig. 9.
The structure of such an automated dispatch control
system contains at least four intelligent agents.
IA1 is a cyclically functioning agent. Based on the results of operational information processing (switch positions), it forms the current switching model of the network and records the events of equipment shutdown and separation of network sections. It uses information about the actuation of relay protection and automatic devices.
IA2 (agent "repair") is initiated at events disconnecting equipment (which are fixed by IA1 and processed in IA4), given (emergency) shutdown requests for the respective items of equipment, and adjustment, where necessary, the appropriate limitations of operating conditions [12, 16-26].
Formalized operating instructions stored in the Knowledge Base of the system are used for automatic operational processing of repair requests. The instructions specify restrictions on the operating parameters (these are normally active power flows in the control cutsets), which must be observed when disconnecting the equipment named in the application. These restrictions can be of two types: those imposed on the duration of the claim and the restrictions on the switching times of switching devices. Inconsistency between the restrictions on the application being processed and the restrictions on the already resolved applications may lead to the refusal of one of the conflicting applications or to the most severe restriction on the time of "imposition" of applications. When repairs are terminated on an already resolved request due to a failure, the emergency readiness time required to change the repair schemes of the corresponding facilities is taken into account. Module IA2 compiles a table of repairs for dispatchers with appropriate restrictions [19-23].
IA3 (agent "Relay protection and automatic control equipment") operates cyclically and detects events actuating relay protection devices and automatic controls of an object.
The IA3 functions, apart from other specifical "relay" tasks, include the formation of events associated with the operation of relay protection and automatic equipment and emergency control devices. The source of this information is the object data concentrators. Since this paper considers
Fig. 10. Differential relay, a) no internal damage, b) internal damage.
h
7:
■///-
■Hh
u2
input layer hidden layer
abs
output layer
abs
ab<t//j) Yipur^
anz^iUI)
«1» or «0» l )——►
WhL
Fig. 11. The structure of the power transformer protection algorithm based on a neural network.
only dispatching express analysis of situations, the events triggering the relay protection and automatic devices (in which the "correct" sequence of events is significant rather than accurately recorded time of the event) have the same status as the events triggering switching devices (recorded in IA1). In the future, the automation of a complete "relay" analysis can make the IA3 functions complicated.
IA4 (agent "Analysis") is initiated when detecting relay operation, equipment shutdowns, separation of network sections. This agent generates a description of the situation and (if necessary) displays advice texts for the dispatcher [23-26].
Diagnostics of electrical equipment faults and overhead power line condition by monitoring systems (Smart Grid) is investigated in [27], while [28-32] focus on ensuring reliable operation of electric networks and experience in testing of digital substation primary equipment and relay protection.
IX. Application of neural network algorithms for the recognition of turn-to-turn faults in power transformers
The application of neural network algorithms to recognize turn-to-turn faults in power transformers is studied in [33]. One of the most difficult problems in protecting power transformers is to detect small-scale internal faults.
The differential protection scheme is based on the principle that the input power of the power transformer, under normal conditions, is equal to the output power. Under normal conditions, no current flows into the current coil of the differential relay. When a fault occurs within the protected area, the current balance is disturbed, the relay contacts close and a signal is sent to certain circuit-breakers to trip the faulty equipment. Differential relay KA compares primary and secondary currents of a power
transformer. Current transformers (TT) are used to reduce the magnitude of primary currents so that their secondary currents are equal.
Figure 10 shows a differential relay in its simplest form. The polarity of the current transformers is chosen such that the current circulates without passing through the relay under normal load conditions and external faults.
Thus, based on the above studies, the most effective structure of the neural network is selected. The structure of the selected configuration is shown in Fig. 11.
The assessment of the operation of differential protection relies on an already created algorithm for the operation of differential protection and a model of power transformer with a turn-to-turn fault [33].
The principles of constructing intelligent relay protection of electrical networks, as well as an algorithm for identifying the damaged section on cable-overhead power lines based on the recognition of wave portraits, are investigated in [34-36].
X. Conclusion
1. Two historically established directions (expert systems and neural networks) of building artificial intelligence systems are studied in detail.
2. The parallels and analogies with the work of the human brain are considered to use the algorithms and principles of the brain to create artificial intelligence.
3. The studies on the creation of a human brain model to be used in the construction of an artificial intelligence system are considered.
4. It is noted that the high performance of information processing in the human brain can only be explained by the parallel operation of many relatively slow neurons and a large number of mutual connections between them.
5. The processes taking place in the human brain cannot be embedded in the coordinate system of the Euclidean
space R3, so one can try to embed them in other types of spaces, for example, multidimensional.
6. It is emphasized that the theoretical foundations of neural networks were laid in the 1940s.
7. The hierarchical structure of the brain is confirmed by the existence of towers of mental spaces, the ability of the brain to simultaneously operate in several mental spaces, i.e., to process in parallel cognitive information of different mental levels.
8. The physical representation of a mental space of hundreds of thousands of mental points requires a Euclidean space of inconceivable dimension. The brain is a result of the projection of a huge number of mental points onto each point of the physical region of the brain. Therefore, an infinite p-adic tree can be isometrically embedded only in an infinite-dimensional Hilbert space.
9. Consciousness represents vortices of events distributed throughout the brain that competes for the brain's attention. Thus, the result appears to us when one of these vortices of consciousness manages to become dominant over all the others, and the brain rationally justifies the result retroactively, and it seems to us that all consciousness is under the control of a single center.
10. Thus, based on the "developments" of the human brain, created over millions of years of evolution, the creation of an artificial intelligence system (we do not have these millions of years) should be guided by several principles:
• the parallel operation of many relatively slow processor chains and a large number of mutual connections between them,
• the hierarchical structure of information flows will make it possible to simultaneously operate in several spaces, i.e., to process information of different levels concurrently,
• the work of such a complex structure as artificial intelligence is possible only as an endless tree, at the top of which there is a guiding center, and below - a pyramid of branches leading to a multitude of less important centers, and which can be isometrically embedded only in an infinite-dimensional space.
11. Attempts can be made to teach (train) neural networks and use them in pattern recognition tasks, strategic games, and others.
12. Expert systems are of priority for "routine" operational tasks in the electric power industry (such as the automated control system for power grids).
13. The system of wireless transmission of data on the leakage current through a surge arrester based on ZigBee technology allows organizing a radio network by uniting all devices into a single multi-mesh network, which is analogous to a neural network.
14. High-quality information support of dispatching decisions in emergencies requires the systems for the dispatching control of electrical networks to be
implemented based on intelligent agents relying on expert systems technology.
15. It is necessary to ensure that the dispatching system is provided with a sufficiently complete volume of tele-signaling of data on the position of switches and data on composition and operation of the relay protection and automatic devices.
16. A sample of an intelligent dispatching system for managing electrical networks has been developed.
17. Analysis of emergencies in a networked intelligent system is the basis for building an intelligent system (a dispatcher's advisor for electrical network companies) and should be performed at two levels: the substation and the electrical network.
18. When building complexes of automated dispatching control systems for electric power systems and power grids, it is advisable to use a multi-agent structure of intelligent functions, with the intelligent agents localized at the workplaces of technologists.
19. "Volumetric" intelligent ADCS in electrical grids should use the architecture of intelligent systems with a "Notice Board" and obtain a computer certificate of a technological breakdown, which will substantially facilitate the subsequent analysis of the accident with the participation of human specialists.
20. The concept of combining intelligent natural language systems with the extreme programming methodology opens up the possibility of relatively simple and effective development of intelligent agents while minimizing the participation of professional programmers in this development.
21. The recognition of internal turn-to-turn faults in the transformer winding has been investigated, and an algorithm for the power transformer protection based on a neural network has been created.
Acknowledgment Cooperation of Universities and Innovation Development, Doctoral School project "Complex diagnostic modeling of technical parameters of power transformer-reactor electrical equipment condition," 2009 has made the publishing of this article possible.
References
[1] Neural networks and expert systems. [Online]. Available: https://mybiblioteka.su/11-93487.htm. (in Russian)
[2] Artificial neural networks and expert systems. [Online]. Available: https://vikidalka.ru/2-178443. html. (in Russian)
[3] Expert systems. Electronic textbook. A course of lectures. [Online]. Available: http://256bit.ru/Expert/ Glava%2023/Index8.htm. (in Russian)
[4] Neural systems and networks. [Online]. Available: https://helpiks.org/4-37479.html. (in Russian)
[5] Neural networks. [Online]. Available: https://
studref.com/336179/informatika/neyronnye_seti. (in Russian)
[6] A. V. Mogilenko, "Artificial intelligence: methods, technologies, energy applications," Russian Energy Agency of the Ministry of Energy of Russia, Russia. Analytical review, 2019. [Online]. Available: https:// in.minenergo.gov.ru/upload/iblock/2f9/2f9942ed690 77878d3421e1ba259312c.pdf. _(in Russian)
[7] A. Yu. Khrennikov, Modeling of thinking processes in p-adic coordinate systems. Moscow, Russia: Publishing house FIZMATLIT, 2004, 296 p., ISBN 5-9221-0501-9.
[8] Z. Freud, Psychology of the unconscious: collection works, M. G. Yaroshevsky, Ed. Moscow, Russia: Prosveshchenie, 1990, 448 p.
[9] O. Shor, A, Glik, A. Yaniv-Rosenfeld, A. Valevski, A. Weizman, A. Khrennikov, et al, "EEG p-adic quantum potential accurately identifies depression, schizophrenia, and cognitive decline," PLoS ONE, vol. 16(8), pp. e0255529, 2021. DOI: 10.1371/journal. pone.0255529.
[10] M. Kaku, The future of the mind. New York, USA: Doubleday, 2014.
[11] R. Carter, How the brain works. Publisher: Corpus, 2014, 224 p.
[12] Yu. Ya. Lyubarsky, Intelligent information systems. Moscow, Russia: Nauka, 1990, 232 p. (in Russian)
[13] Yu. Ya. Lyubarsky, A. Yu. Khrennikov, The use of elements of artificial intelligence: computer support for operational solutions in smart power grids. Study guide. Litres, 2021, 140 p. [Online]. Available: https://www.litres.ru/aleksandr-urevich-hr/ ispolzovanie-elementov-iskusstvennogo-intellekta-komp (in Russian).
[14] M. B. Berkinblit, Neural networks. Moscow, Russia: Miros and VZMSh. RAO, 1993, 96 p. (in Russian)
[15] OJSC "Positron", Remote control system of leakage current of surge arrester ("Ariadna"). [Online]. Available: http://docplayer.ru/66507440-0ao-pozitron-sistema-distancionnogo-kontrolya-toka-utechki-opn-kv-udktu-opn-ariadna.html_(in Russian)
[16] A. I. Khalyasmaa, P. V. Matrenin, S. A. Eroshenko, "Errors analysis for the application of machine learning algorithms in the power industry," Electric Power. Transmission and distribution, no. 3 (66), pp. 46-53, 2021. (in Russian)
[17] Yu. Ya. Lyubarsky, V. D. Vlasyuk, S. K. Kakovskiy, "Automation of search for faults in electrical distribution networks," Electric stations, no. 4, pp. 29-36, 2015. (in Russian)
[18] Yu. Ya. Lyubarsky, S. V. Anashkin, S. K. Kakovsky, S. V. Kartashev, A. G. Miroshkin, "Simulator for dispatching personnel of electrical network," Electric stations, no. 2, pp. 2-6, 2014. (in Russian)
[19] Yu. Y. Lyubarsky, "Intelligent reasoning models in the electrical network dispatcher advisor," Electric stations, no. 3, pp. 35-39, 2017. (in Russian)
[20] A. Yu. Khrennikov, Yu. Y. Lyubarsky, "Computer support for dispatching solutions in smart grids," Electrical Engineering Library, no. 8. pp. 1-92, 2019. (in Russian)
[21] A. Yu. Khrennikov, Yu. Y. Lyubarskiy, "Intelligent systems with a bulletin board for automated dispatch control systems for electrical networks," News of Elektrotechnics, no 3 (117), pp. 34-40, 2019. (in Russian)
[22] A. Yu. Khrennikov, N. M. Alexandrov, Yu. Y. Lyubarsky, "Operational management of power plants and the Internet of Things. Problems of digitalization,"
Electrical Engineering News, no. 4-5. pp. 22-26,
2019. (in Russian)
[23] A. Yu. Khrennikov, Yu. Y. Lyubarsky, N. M. Alexandrov, "Operational management of power grids. Artificial Intelligence Platforms," News of Electrotechnics, no. 6 (120), pp. 30-33, 2019. (in Russian)
[24] A. Yu. Khrennikov, Yu. Y. Lyubarsky, N. M. Alexandrov, "Intelligent power grid management systems. Explanation function," News of Electrotechnics, no. 1 (121) - 2 (122), pp. 50-52,
2020. (in Russian)
[25] A. Yu. Khrennikov, Yu. Y. Lyubarsky, "Distributed and mobile energy storage in networks. Digital control technologies," News of Electrotechnics, no. 4 (124) - 5 (125), pp. 62-64, 2020. (in Russian)
[26] A. Yu. Khrennikov, Yu. Y. Lyubarsky, "Analysis of emergency situations in power grids. Simulator for the dispatcher," News of Electrotechnics, no. 6 (126), pp. 24-26, 2020. (in Russian)
[27] A. Yu. Khrennikov, Diagnostics of Electrical Equipment Faults and Power Overhead Transmission Line Condition by Monitoring Systems (Smart Grid): Short-Circuit Testing of Power Transformers. New York, USA: Nova Publishers, 2016, 188 p. [Online]. Available: https://novapublishers.com/ shop/diagnostics-of-electrical-equipment-faults-and-power-overhead-transmission-line-condition-by-monitoring-systems-smart-grid-short-circuit-testing-of-power-transformers/
[28] A. Yu. Khrennikov, A. A. Kuvshinov, I. A. Shkuropat, Providing reliable operation of electric networks. New York, USA: Nova Publishers, 2019, 296 p. [Online]. Available: https://novapublishers.com/shop/ providing-reliable-operation-of-electric-networks/
[29] A. Y. Khrennikov, N. M. Aleksandrov, "Approaches and Experience in Testing of Digital Substation Primary Equipment and Relay Protection," in Proc. the 10th International Scientific Symposium on Electrical Power Engineering (Elektroenergetika 2019), Technical University of Kosice, Stara Lesna, Slovakia, Sep. 2019.
[30] A. Yu. Khrennikov, T. V. Ryabin, N. M. Aleksandrov, "Diagnostic methods for detection of electrical equipment's faults, defects," in International Conference on Condition Monitoring, Diagnosis,
and Maintenance, CMDM 2017, (4th edition), Sep. 25-27, 2017.
[31] A. Yu. Khrennikov, N. M. Aleksandrov, I. N. Zhelbakov, T. A. Khrennikova, "Smart Grid for increasing the reliability of electric power supply of green cities by detecting electrical equipment faults and defects," in 2nd International Conference on the Applications of Information Technology in Developing Renewable Energy Processes & Systems (IT-DREPS), Petra University, Amman, Jordan, Dec. 6-7, 2017.
[32] A. Yu. Khrennikov, N. M. Aleksandrov, P. S. Radin, "Dependability of Service of Substation Electrical Equipment: Estimation of the Technical Condition State with the Use of Software and Information Tools," in Proc. the Fourteenth International Conference on Dependability of Computer Systems DepCoS-RELCOMEX "Engineering in Dependability of Computer Systems and Networks," Jul. 1-5, 2019, Brunow, Poland, pp. 274-283.
[33] A. L. Kulikov, A. N. Sovina, "Application of neural network algorithms for the recognition of turn-to-turn short circuits in power transformers." [Online]. Available: https://drive.google.com/drive/ folders/1U3px3oVdxodl7rALUourALdWwTc-4RcW?usp=sharing.
[34] V. F. Lachugin, D. I. Panfilov, A. L. Kulikov, A. A. Ryvkin, M. D. Obalin, "Principles of constructing intelligent relay protection of electrical networks," Proceedings of the Russian Academy of Sciences. Power Engineering, no. 4, pp. 28-37, 2015. (in Russian)
[35] A. L. Kulikov, D. I. Bezdushny, M. V. Sharygin, V. Yu. Osokin, "The use of the support vector machine in the implementation of multidimensional relay protection," in Methodological Issues in Reliability Study of Large Energy Systems: Vol. 71. Methodological and practical problems of reliability of energy systems", book 2, N. I. Voropai, Ed. Irkutsk: ESI SB RAS, 2019. pp. 25-34. (in Russian)
[36] A. L. Kulikov, A. A. Loskutov, P. S. Pelevin, "Algorithm for the identification of a damaged area on cable-overhead power lines based on the recognition of wave portraits," Electricity, no. 3, pp. 11-17, 2018. (in Russian)
Alexander Khrennikov was born in Bratsk, Russia, in 1964. He received a Ph.D. degree in Electrical Engineering from Samara City University of Technology in 2009. Currently, he works as Scientific Secretary of Scientific and Technical Center of Federal Grid Company of Unified Energy System, Russia. A.Yu. Khrennikov is an author of more than 250 scientific and technical publications. His main research interests include artificial intelligence systems, expert systems, monitoring the state of electrical equipment, operational dispatch control of power grids, emergency situations, automated dispatch control systems, transformer short-circuit testing, transformer winding fault diagnostic, frequency Response Analysis, smart grid, and information-measuring systems. He is a Distinguished Member of CIGRE and a Professor of Samara Technical State University, Russia..
Yuri Lyubarsky was born in 1938, graduated from Moscow Power Engineering Institute in 1961. Since 1961, he has worked at the All-Union Scientific Research Institute of Electric Power Industry as the head of the Laboratory of Expert Systems. He received a Ph.D. degree in 1965 and a D.Sc. degree in 2000. At present, Yu.Ya. Lyubarsky is the Chief Researcher of the Scientific and Technical Center of Federal Grid Company of Unified Energy System, Russia. He has more than 100 published works, including 30 copyright certificates of the USSR and patents of the Russian Federation, 3 monographs, and 2 books. His research interests include artificial intelligence, expert systems for the electric power industry, computer assistance in the dispatch control of electrical networks and systems.
Andrei Khrennikov was born in 1958 in Volgograd and spent his childhood in the town of Bratsk, in Siberia. In 1975-1980, he studied at Moscow State University, Department of Mechanics and Mathematics. He received a Ph.D. degree in 1983 and a D.Sc. degree in physics and mathematics from the
Department of Mathematical Physics, Steklov Mathematical Institute, Russian Academy of Sciences, in 1990. At present, he is a professor of applied mathematics and the director of the International Centre for Mathematical Modelling in Physics, Engineering, Economics, and Cognitive Science, and Head of Research of Mathematical Institute of Linnaeus University, Vaxjo Sweden. Prof. Khrennikov published 714 articles in mathematical, physical, and biological journals with 6020 citations; his H-index is 36. He published 20 monographs (Cambridge Univ. Press, Oxford Univ. Press, Springer, World Scientific, FizMatlit, Nauka).
