CC BY
15
DOI: 10.24412/2413-2527-2022-432-85-93
Neurointerface Modeling for Controlling Dynamic Systems
Yu. V. Tsymay, Grand PhD A. P. Nyrkov, M. V. Kardakova Admiral Makarov State University of Maritime and Inland Shipping Saint Petersburg, Russia [email protected], [email protected], [email protected]
Abstract. Modeling of input parameters for systems controlled by neural interfaces is considered. The relevance of the issue for transport technological systems is considered. Bioelectric signals are used as input parameters for activating control code algorithms. When forming the input signal, the parameters of the rhythms of the electroencephalogram are taken into account. The result is a device capable of reading the electromyogram signals and acting on the control object in accordance with a given algorithm. The input signal from the sensors varies depending on the condition of the muscles to which the electrodes are connected. The signal controls the translational and rotational movements of the layout. The software control model is based on the principles of flexible logic, so it is possible to configure the system for biological signals of varying accuracy.
Keywords: programming of neural interfaces, control code, bioelectric signal, microcontroller.
Introduction
Technical systems with built-in neural interfaces have appeared in various industries for a long time. Initially, the technology of neurointerfaces was used in medicine to identify various disorders of the brain or for its research. But scientific research in this area has proven the effectiveness of the use of neurotechnologies for controlling industrial manipulators and robots, as well as various automated systems. The principle of neural networks expands control capabilities and increases the number of algorithms for interaction within systems. These are so-called artificial intelligence systems in the field of business, medicine [1], geology, physics [2], transport, etc. [3].
When designing and creating neural interfaces, scientific knowledge of interdisciplinary fields such as biology and human physiology, physics, mathematics, programming, circuit engineering is used. Complex hardware and information systems connect impulses of brain neurons or signals coming from contracting muscles and external control objects. Management objects can be physical and virtual [4].
When creating neural interfaces, the following are used:
• standard algorithms and diagrams for processing electrical signals of the human cerebral cortex;
• software testing algorithms;
• methods of forming technical specifications and drawing up electrical functional circuits;
• the principle of operation of the brain-computer interface;
• visualization methods and communication with a programmable logic controller;
• setting limit input values;
• technologies of software and industrial interfaces «brain-computer»
The main task at the initial stage of designing neural interfaces is to create algorithms for recognizing the main rhythms of the brain and muscle contractions. If the input parameters are misinterpreted, the technological system will start processing an inaccurate or false signal. This can lead to an incorrect output signal and disrupt its safe operation [5, 6].
In order to avoid such situations, it is necessary to fine-tune special software and peripheral device drivers, if necessary.
A person is a complex object of measurement. It is necessary to take into account its psychological and physiological characteristics. Only then can the measurement results be used as an input source for the technical system. And understanding the fundamental differences of biological signals will allow you to choose the correct processing algorithm [7].
It is important to know and use in analytical calculations the features of connection and operation of devices for recording and analyzing electroencephalograms and evoked potentials. The level of accuracy of the measuring device gives a signal of appropriate quality. The amplitude characteristics of the electroencephalogram create the initial data for the formation of the input signal and the elimination of artifacts. Artifacts appear when a signal is received from the electrodes to the input ports. The interface device converts the input signal and generates an input control signal, but interference occurs. These are the artifacts. You can remove them only through a handler program or through improving the interface device.
When forming the input signal, the parameters of the rhythms of the electroencephalogram are taken into account. This allows you to choose a method for processing and analyzing the measured electroencephalogram signal and evoked potentials [8].
Oscilloscope, tester, potentiometer, rheostat are used as measuring instruments.
The design of the software architecture of industrial interfaces «brain-computer» is carried out in C++, Python. Other options are possible. The choice of language largely depends on the controlling microprocessor device [9, 10].
The microprocessor devices currently in use are capable of receiving an analog signal converted into a digital one. The signal type is changed by means of analog-to-digital converters built into the architecture of the control device or used as additional modules.
The block diagram of the technological system allows to obtain the initial nomenclature list for modeling the specified devices of industrial interfaces «brain-computer». The computational model will allow the analysis of physical processes occurring in electronic telecommunication devices. Modeling of the software and hardware of the technological system will
determine the main points of the system in which signal amplification or filtering is required, the need to adjust signal processing algorithms from the cerebral cortex or from muscle contractions [11-13].
The mathematical model takes into account the choice of the method of preparation and analysis of the psychological and emotional state of the measured object. The correct choice of the method affects the formation of the right attitude of the operator and determines the quality and accuracy of measurement.
Modeling of the brain-computer interface
Modeling stages
The main stages of modeling the brain-computer interface will be the following:
1. Assembly and adjustment of a bionic movable layout: selection of the elementary base of components, assembly of the mechanical part of the layout, installation and connection of electronic components, adjustment of the layout to perform elementary translational and rotational movements.
2. Creation of program code to check the autonomous operation of the layout: building a block diagram of the device, forming a control code for an electrical circuit based on microprocessor electronics.
3. Investigation of the bioelectric activity of the operator's muscles using sensors to take an electromyogram: the use of psychological and physiological methods of adjustment and correction of the measuring object, the use of measuring instruments.
4. Electromyogram visualization: checking the visualization from each sensor, recording and processing the signal from the synchronization device.
5. Carrying out a functional test for the registration of artifacts.
6. Registration of an electromyogram during jaw compression.
7. Registration of an electromyogram when blinking.
8. Analysis of the influence of interference on the electroencephalogram signal: signal processing through special software, the final configuration of the circuit and control logic.
9. Creation of program code for signal processing and control of I/O ports.
10. Creation of an autonomous control system for a bionic model of a human hand.
11. Implementation of the possibility of configuring the system for a specific operator: checking the control of hand movements using electromyogram and electroencephalogram signals coming from the measuring object under various muscle conditions [14].
Hardware modeling and measurement
of input parameters
within the framework of hardware modeling, a bionic model with the number of degrees of freedom necessary for transla-tional and rotational motion was assembled and configured. The selection of parts and electronic components for the mechanical part of the layout was carried out from a standard list.
The Arduino microcontroller board is used as the control unit. The signal is transmitted to the system via an expansion board. The servo drive ensures the mobility of the model elements. After assembly and connection, the layout is configured:
• determination of minimum and maximum shaft deflection angles for all servos;
• determination of the direction and speed of rotation.
To implement the correct configuration of the layout, additional electronic components are used: potentiometers, LEDs. The first ones provide mechanical debugging of the trajectory. The second one checks the correctness of the logical assembly of the hardware model and the correctness of the connection of the i/o ports. This is important if there is no main control code yet [15].
Electromyogram sensors are connected to the Arduino board to study the bioelectric activity of the muscles of the object. Electromyograms from each sensor are visualized. individual elements are detected in the signal and processed independently. standard software provided by the Arduino board developer company is used for processing.
The result of the described process is a device capable of reading the electromyogram signals and acting on the control object in accordance with a given algorithm.
The next task is to obtain a single autonomous system that combines a bionic mobile layout and a control device with elec-tromyogram sensors. The input signal from the sensors varies depending on the condition of the muscles to which the electrodes are connected. The signal controls the translational and rotational movements of the layout. The number of states depends on the trajectories of the moving parts of the model [16].
The control program is based on the principles of flexible logic [17]. Therefore, the system can be tuned to biological signals of varying accuracy [18]. The source of the biological signal can also be adjusted and changed.
under ideal experimental conditions, the same type of biological signals received from different sources affect the control object identically.
The source of the input signal can be not only the bioelectric activity of the muscles, but also a-waves measured using electrodes fixed on the surface of the operator's head. Also, these electrodes receive a signal initiated by blinking or jaw compression. such measurement requires higher accuracy in processing. This is due to the high probability of the appearance of many artifacts during signal registration.
Visualization of input parameters
Visualization of input parameters was carried out using the Bitronics Studio software. Fifty comparative measurements were performed. of these, the first ten can be considered as installation ones, since the electrode contact was adjusted and the search for an equilibrium psychological and physical state of the measuring object was performed. The results of each measurement are analyzed for the appearance of artifacts. The control signal is formed from the most accurate measurement results. The fewer artifacts, the better.
Figure 1 shows the results of measuring and processing the electromyogram signal from one sensor, taking into account the identified artifacts. The electrode contacts the relaxed muscles.
Figure 2 shows the results of measurements with the same electrode without changing the point of contact, but with muscle tension.
Then, before processing the electromyogram signal, a program code was created to control the actuators. An example of the code is shown in Figure 3 [14].
Рис Маркер
Х0: 5.000 XI: 15.00 N: 190.0
Y0: 2.529 VI: 2.510 A: 2.506
STD: 0.037
MAX: 2.6(
MIN: 2.43
эмг
0 3 6 9 12 15 18
Time, s
Fig. 1. Visualization of an electromyogram from a single sensor. Muscles are relaxed
Рис Маркер I
XO: 5.000 YO: 2.490
XI: Yl:
15.00
1.255
N:
A:
STD: MAX: M1N:
192.0 2.401 0.65f 4.9<; o.3i:
эмг
Fig. 2. Visualization of an electromyogram from a single sensor. Muscles are tense
13
14 void loop () {
15 emg1=analogRead (AO);
16 Serial.write ("AO"};
17 II Serial.write {emgl);
18 sign1=map(emg1, 0,1023, 0,255);
19 Serial.write (signl);
20 if (sign1>153)
21 {
22 digitalWrite (LEDpinl, HIGH);
23 }
24 else
25 {
26 digitalWrite (LEDpinl, LOW);
27 }
28
29 emg2=analogRead (A1);
30 delay (50);
31 sign2=map(emg2, 0,1023, 0,255);
32 Serial.write (sign2);
33 if (sign2>135)
34 {
35 digitalWrite (LEDpin2, HIGH);
36 }
37 else
38 {
39 digitalWrite (LEDpin2, LOW);
40 }
= I
сиетч использует 2266 Сайт <7*| памяти устройства, всего поступив 32256 Оайт. ьные переменные используют 196 С-айт <9%) динамической памяти, оставляя 1952 Байт для локальных ксимум: 2049 сайт.
Fig. 3. The program code for controlling the actuators before processing the electromyogram signal
The program for signal processing (finding the amplitude) and controlling the actuators using the processed electromyogram signal is shown in Figure 4.
The software and hardware parts of the model are combined. the input signal is corrected. The measurement results are loaded into the handler. The signal is filtered and compiled.
Figure 5 shows the visualization of the processed signal with relaxed muscles.
Figure 6 shows the visualization of the processed signal with tense muscles.
10 11 12
13
14
15
16
17
18
19
20 21 22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
int sign2MlN=255; int sign1AMPL=0; int sign2AMPL =0: int ¡=0;
void aa() {;
sign1AMPL=0;
sign2AMPL =0;
for (i; i<36; i++) {
emg1 =arialogRead(A0);
stgn1=map{emg1, 0, 1023, 0, 255);
emg2=analogRead(A1 );
stgn2=map(emg2, 0, 1023, 0, 255);
if (sign 1 > signlMAX) sign1MAX= signl if (signl > signlMIN) sign1MIN= signl; if (sign2> sign2MAX) sign2MAX= sign2: if (sign2> sign2MIN) sign2MIN= sign2;
}
sign 1 AMPL=0.3*sign 1 +0,7*( signlMAX- signlMIN); sign2AMPL=0.3*sign2+0,7*( sign2MAX- sign2MIN); Serial.write("A0"); Seriai.write (signlAMPL);
gn1MAX=0: gn1MIN=255: gn2MAX=0: gn2MIN=255: ¡=0;
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 61 62
63
64
65
66 67
void setupO {;
Serial.begin (9600); pinMode (LEDpinl .OUTPUT); pinMode {LEDpin2,OUTPUT);
}
void loopO {; aa ();
if (signl >60);
digitalwrite(LEDpin1,H!GT);
}
else
{
{
digitalwrite(LEDpin1,LOW);
if (sign2>60);
}
else
{
}
digitalwrite(LEDpin2,HIGT);
digitalwrite(LEDpin2,LOW);
Скетч использует 3334 Оайт (10%) памяти устройства. Всего досту Глобальные переменные используют 206 Сайт (101) динамической па скетч использует 3334 сайт |îû%) гаилти устройства, всего доступно 32256 '
Fig. 4. A program for signal processing and control of actuators using a processed electromyogram signal
Fig. 5. Visualization of the processed signal with relaxed muscles
Fig. 6. Visualization of the processed signal with tense muscles
The processed signal is used as the initial information to activate the control algorithm for dynamic processes in the model under consideration. In this case, when the muscle is strained by the operator, the model makes a linear forward movement, when the muscle relaxes, it stops. A different degree of tension provokes a forward backward movement or rotation of the moving part of the model clockwise and counterclockwise.
Thus, a control code has been created to influence the technical system in five parameters. The operator must maintain the concentration of attention selected during the preliminary tests. It reproduces the same level of muscle tension. Only then does the accuracy and correctness of the reaction of the mechanical part of the model meet expectations.
When using two sensors, the following algorithm can be set for the control program.
Depending on the tension and relaxation of the two muscles operator (for example, the muscles, the flexors and extensors in the forearm) are translational, rotational, and reverse motion: momentum (both muscles tense), moving backward (both muscles are relaxed), clockwise rotation (muscles flexors tight extensors are relaxed), counterclockwise rotation (muscles relaxed flexors, extensors tense).
In the program code, a library is connected to work with servos. The minimum and maximum rotation angle for each servo is set. The speed of movement of the moving elements of the model is set. Software objects are created to work with each servo motor.
Variables are formed to store the raw and processed values of the electromyogram. Threshold values are indicated for determining the state of the muscle — relaxed/tense.
Variables for storing minimum and maximum values during processing are specified in the range from 0 to 255. A memory area is allocated to store the rotation angle of the servo shaft (for each servo).
To calculate the amplitude of the electromyogram, a subroutine is used [19]:
void calc_amp() {
for (int k = 0; k < 32; k++) {
emgl = analogRead(AO);
emg2 = analogRead(Al);
emgl = map(emg1, 0, 1023, 0, 255);
emg2 = map(emg2, 0, 1023, 0, 255);
if (emgl > maxl)
maxl = emgl;
if (emgl < minl)
minl = emgl;
if (emg2 > max2)
max2 = emg2;
if (emg2 < min2)
min2 = emg2; }
ampl = 0.3*ampl + 0.7*(maxl - minl); amp2 = 0.3*amp2 + 0.7*(max2 - min2);
The main code describes the main events in the system, taking into account the degree of muscle tension, conditions for reaching thresholds or inactivity.
Conclusion
The use of neural interfaces to control production and transport systems will improve the accuracy of the control code thanks to methods and algorithms for processing and converting the original signals obtained by measuring the bioelectric activity of muscles and alpha waves of the brain [20, 21]. Increasing the level of information security depends on the limitations of the model. Constraints are influenced by external factors that initiate dynamic processes [21-24].
However, the disadvantage can be considered laborintensive when changing the biosignal source. In this case, an adjustment of the input code is required, because according to physiological and psychological indicators, bioelectric signals from different objects cannot have a complete match. This leads to the question of interpretation of such signals. Current-
ly, there are no precise algorithms and techniques that give an indisputable reading of bioelectric parameters. The quality largely depends on the features of the signal visualization software, the accuracy of measuring instruments, and the quality of the connecting electrodes.
Also, a number of uncertain parameters arise during the mathematical analysis of dynamic processes in industrial and transport systems [21]. In this regard, forecasting the development of events causes difficulty and generates many options. This forces developers to create a multi-level control program that includes algorithms for as many probabilities of possible states of the system as possible. For a typical dynamic system with certain parameters, this probability is a value having from ten to fifty digits. In technical systems controlled by neural interfaces, the number of elementary states can lead to redundancy of the control code. It will take too many probabilistic algorithms. In the future, it is necessary to develop a software and mathematical apparatus capable of efficiently analyzing input parameters with a large number of artifacts. Then you can use weaker signals.
Acknowledgements
The authors express their gratitude to the chief expert of the regional championships in St. Petersburg World Skills Russia «Designing neural interfaces» Ilyushina A. N. for advice on the issue under study.
References
1. Sokolov S. S., Kondratiev A. N., Lesteva N. A., Diyagi-na N. V. Automation of Intraoperative Analysis of Indicators the Inflammatory Response of Neurosurgical Patients Undergoing Brain Tumors Removal, Proceedings of the Ural Environmental Science Forum «Sustainable Development of Industrial Region» (UESF-2021), Chelyabinsk, Russia, February 17-19, 2021. E3S Web of Conferences, 2021, Vol. 258, Art. No. 04005, 10 p. DOI: 10.1051/e3sconf/202125804005.
2. Sokolov S. S., Zhilenkov A. A., Chernyi S. G., et al. Hybrid Neural Networks in Cyber Physical System Interface Control Systems, Bulletin of Electrical Engineering and Informatics, 2020, Vol. 9, No. 3, Pp. 1268-1275.
DOI: 10.11591/eei.v9i3.1293.
3. Aggarwal C. C. Neural Networks and Deep Learning: A Textbook. Cham, Springer Nature, 2018, 520 p.
DOI: 10.1007/978-3-319-94463-0.
4. Altaisky M. V., Kylov V. A. Signal Identification Based on Multiscale Decompositions. In: Sree Hari Rao V. (ed.). Information Technology for Real World Problems. Hyderabad, Orient Blackswan, 2011, Pp. 178-221.
5. Bagnall A., Lines J., Bostrom A., et al. The Great Time Series Classification Bake Off: A Review and Experimental Evaluation of Recent Algorithmic Advances, Data Mining and Knowledge Discovery, 2017, Vol. 31, Is. 3, Pp. 606-660. DOI: 10.1007/s10618-016-0483-9.
6. Goodfellow I. NIPS 2016 Tutorial: Generative Adversarial Networks, arXiv, 2017, Vol. 1701.00160, 57 p.
DOI: 10.48550/arXiv.1701.00160.
7. Foster D. Generative Deep Learning: Teaching Machines to Paint, Write, Compose and Play. Sebastopol (CA), O'Reilly Media, 2019, 330 p.
8. Esbensen K. H., Guyot D., Westad F., Houmoller L. P. Multivariate Data Analysis — In Practice. Fifth Edition. An In-
troduction to Multivariate Data Analysis and Experimental Design. Oslo, CAMO Software AS, 2004, 616 p.
9. McKinney W. Python for Data Analysis: Data Wrangling with Pandas, NumPy, and IPython. First Edition. Sebastopol (CA), O'Reilly Media, 2012, 466 p.
10. Grolemund G., Wickham H., R for Data Science: Import, Tidy, Transform, Visualize, and Model Data. First Edition. Sebastopol (CA), O'Reilly Media, 2017, 518 p.
11. Specht D. F. A General Regression Neural Network, IEEE Transactions on Neural Networks, 1991, Vol. 2, Is. 6, Pp. 568-576. DOI: 10.1109/72.97934.
12. Srivastava N., Hinton G. E., Krizhevsky A., et al. Dropout: A Simple Way to Prevent Neural Networks from Overfitting, Journal of Machine Learning Research, 2014, Vol. 15, Pp. 1929-1958. DOI: 10.5555/2627435.2670313.
13. Goodfellow I. J., Pouget-Abadie J., Mirza M., et al. Generative Adversarial Nets, arXiv, 2014, Vol. 1406.2661, 9 p. DOI: 10.48550/arXiv.1406.2661.
14. Sutskever I., Martens J., Dahl G. E., Hinton G. E. On the Importance of Initialization and Momentum in Deep Learning, Proceedings of the 30th International Conference on Machine Learning (ICML 2013), Atlanta, GA, USA, June 17-19, 2013. Proceedings of Machine Learning Research, 2013, Vol. 28, No. 3, Pp. 1139-1147.
15. Rivero D., Gestal M., Rabuñal J. R. Genetic Programming: key concepts and examples: A brief tutorial on Genetic Programming. Saarbrücken, Lambert Academic Publishing, 2011, 56 p.
16. Nyrkov A. P., Kardakova M. V., Kolesnichenko S. V., et al. Modeling the Operating Range of the Fire Safety System Response Parameters on Board, Proceedings of the 2020 IEEE Conference of Russian Young Researchers in Electrical and Electronic Engineering (ElConRus), Saint Petersburg, Moscow, Russia, January 27-30, 2020. Institute of Electrical and Electronics Engineers, 2020, Pp. 434-437.
DOI: 10.1109/EIConRus49466.2020.9038955.
17. Nyrkov A. P., Zhilenkov A. A., Sokolov S. S., Chernyi S. G. The Use of Fuzzy Control Methods for Evaluation of Complex Systems on the Example of Maritime Fleet Equipment, Vibroengineering PROCEDIA, 2016, Vol. 8, Pp. 6-10.
18. Boriev Z. V., Sokolov S. S., Nyrkov A. P., Nekrasova A. A. Mathematical and Information Maintenance of Biometric Systems, Proceedings of the International Conference on Mechanical Engineering, Automation and Control Systems 2015 (MEACS 2015), Tomsk, Russia, December 01-04, 2015. IOP Conference Series: Materials Science and Engineering, 2016, Vol. 124, Art. No. 012046, 4 p.
DOI: 10.1088/1757-899X/124/1/012046.
19. Svetlov V. A., Dolenko S. A. Development of the Algorithm of Adaptive Construction of Hierarchical Neural Network Classifiers, Optical Memory and Neural Networks (Information Optics), 2017. Vol. 26, No. 1, Pp. 40-46.
DOI: 10.3103/S1060992X17010076.
20. Yemelyanov V. A., Yemelyanova N. Yu., Shved E. V., et al. Modeling of the Multilayer Perceptrons for Image Recognition of the Steel Microstructures, Proceedings of the 2020 IEEE Conference of Russian Young Researchers in Electrical and Electronic Engineering (ElConRus), Saint Petersburg, Moscow, Russia, January 27-30, 2020. Institute of Electrical and Electronics Engineers, 2020, Pp. 952-955.
DOI: 10.1109/EIConRus49466.2020.9038971.
21. Sokolov S. S., Zhilenkov A. A., Chernyi S. G., et al. Dynamics Models of Synchronized Piecewise Linear Discrete Chaotic Systems of High Order, Symmetry, 2019, Vol. 11, Is. 2, Art No. 236, 12 p. DOI: 10.3390/sym11020236.
22. Kotenko I. V., Saenko I. B., Lauta O. S. Analytical Modeling and Assessment of Cyber Resilience on the Base of Stochastic Networks Conversion, Proceedings of the 10th International Workshop on Resilient Networks Design and Modeling (RNDM 2018), Longyearbyen, Norway, August 27-29, 2018. Institute of Electrical and Electronics Engineers, 2018, Pp. 1-8. DOI: 10.1109/RNDM.2018.8489830.
23. Sokolov S. S., Alimov O. M., Trubacheva Yu. A., et al. Providing Security for Virtual Infrastructures Based on an Integrated Mechanism, Proceedings of the 2018 IEEE Confer-
ence of Russian Young Researchers in Electrical and Electronic Engineering (ElConRus), Moscow, Saint Petersburg, Russia, January 29-February 01, 2018. Institute of Electrical and Electronics Engineers, 2018, Pp. 132-135. DOI: 10.1109/EIConRus.2018.8317047.
24. Saenko I. B., Ageev S. A., Kotenko I. V. Detection of Traffic Anomalies in Multi-Service Networks Based on a Fuzzy Logical Inference. In: Badica C., et al. (eds.) Intelligent Distributed Computing X (IDC2016): Proceedings of the 10th International Symposium on Intelligent Distributed Computing, Paris, France, October 10-12, 2016. Cham, Springer Nature, 2017, Pp. 79-88. (Studies in Computational Intelligence, Vol. 678). DOI: 10.1007/978-3-319-48829-5_8.
DOI: 10.24412/2413-2527-2022-432-85-93
Управление техническими системами с помощью нейроинтерфейса
Ю. В. Цымай, д.т.н. А. П. Нырков, М. В. Кардакова Государственный университет морского и речного флота имени адмирала С. О. Макарова
Санкт-Петербург, Россия [email protected], [email protected], [email protected]
Аннотация. Рассматривается моделирование входных параметров для систем, управляемых нейроинтерфейсами. Рассмотрена актуальность проблемы для транспортных технологических систем. Биоэлектрические сигналы используются в качестве входных параметров для активации алгоритмов управляющих кодов. При формировании входного сигнала учитываются параметры ритмов электроэнцефалограммы. Результатом является устройство, способное считывать сигналы электромиограммы и воздействовать на объект управления в соответствии с заданным алгоритмом. Входной сигнал от датчиков варьируется в зависимости от состояния мышц, к которым подключены электроды. Сигнал управляет поступательными и вращательными движениями макета. Модель программного управления основана на принципах гибкой логики, поэтому можно настроить систему на биологические сигналы различной точности.
Ключевые слова: программирование нейроинтерфейсов, управляющая программа, биоэлектрический сигнал, микроконтроллер.
Литература
1. Automation of Intraoperative Analysis of Indicators the Inflammatory Response of Neurosurgical Patients Undergoing Brain Tumors Removal / S. S. Sokolov, A. N. Kon-dratiev, N. A. Lesteva, N. V. Dryagina // Proceedings of the Ural Environmental Science Forum «Sustainable Development of Industrial Region» (UESF-2021) (Chelyabinsk, Russia, 17-19 February 2021). E3S Web of Conferences. 2021. Vol. 258. Art. No. 04005. 10 p.
DOI: 10.1051/e3sconf/202125804005.
2. Hybrid Neural Networks in Cyber Physical System Interface Control Systems / S. S. Sokolov, A. A. Zhilenkov, S. G. Chernyi, [et al.] // Bulletin of Electrical Engineering and Informatics. 2020. Vol. 9, No. 3. Pp. 1268-1275.
DOI: 10.11591/eei.v9i3.1293.
3. Aggarwal, C. C. Neural Networks and Deep Learning: A Textbook. — Cham: Springer Nature, 2018. — 520 p. DOI: 10.1007/978-3-319-94463-0.
4. Altaisky, M. V. Signal Identification Based on Mul-tiscale Decompositions / M. V. Altaisky, V. A. Krylov // Information Technology for Real World Problems / V. Sree Hari Rao (ed.). — Hyderabad: Orient Blackswan, 2011. — Pp. 178-221. — (Universities Press Series in Systems, Models, Informatics and Control).
5. The Great Time Series Classification Bake Off: A Review and Experimental Evaluation of Recent Algorithmic Advances / A. Bagnall, J. Lines, A. Bostrom, [et al.] // Data Mining and Knowledge Discovery. 2017. Vol. 31, Is. 3. Pp. 606-660. DOI: 10.1007/s10618-016-0483-9.
6. Goodfellow, I. NIPS 2016 Tutorial: Generative Adversarial Networks // arXiv. 2017. Vol. 1701.00160. 57 p.
DOI: 10.48550/arXiv.1701.00160.
7. Foster, D. Generative Deep Learning: Teaching Machines to Paint, Write, Compose and Play. — Sebastopol (CA): O'Reilly Media, 2019. — 330 p.
8. Multivariate Data Analysis — In Practice. Fifth Edition. An Introduction to Multivariate Data Analysis and Experimental Design / K. H. Esbensen, D. Guyot, F. Westad, L. P. Hou-moller. — Oslo, [et al.]: CAMO Software AS, 2004. — 616 p.
9. McKinney, W. Python for Data Analysis: Data Wrangling with Pandas, NumPy, and IPython. First Edition. — Sebastopol (CA): O'Reilly Media, 2012. — 466 p.
10. Grolemund, G. R for Data Science: Import, Tidy, Transform, Visualize, and Model Data. First Edition / G. Grolemund, H. Wickham. — Sebastopol (CA): O'Reilly Media, 2017. — 518 p.
11. Specht, D. F. A General Regression Neural Network // IEEE Transactions on Neural Networks. 1991. Vol. 2, Is. 6. Pp. 568-576. DOI: 10.1109/72.97934.
12. Dropout: A Simple Way to Prevent Neural Networks from Overfitting / N. Srivastava, G. E. Hinton, A. Krizhev-sky, [et al.] // Journal of Machine Learning Research. 2014. Vol. 15. Pp. 1929-1958. DOI: 10.5555/2627435.2670313.
13. Generative Adversarial Nets / I. J. Goodfellow, J. Pou-get-Abadie, M. Mirza, [et al.] // arXiv. 2014. Vol. 1406.2661. 9 p. DOI: 10.48550/arXiv.1406.2661.
14. On the Importance of Initialization and Momentum in Deep Learning / I. Sutskever, J. Martens, G. E. Dahl, G. E. Hinton // Proceedings of the 30th International Conference on Machine Learning (ICML 2013) (Atlanta, GA, USA, 17-19 June 2013). Proceedings of Machine Learning Research. 2013. Vol. 28, No. 3. Pp. 1139-1147.
15. Rivero, D. Genetic Programming: key concepts and examples: A brief tutorial on Genetic Programming / D. Rive-ro, M. Gestal, J. R. Rabuñal. — Saarbrücken: Lambert Academic Publishing, 2011. — 56 p.
16. Modeling the Operating Range of the Fire Safety System Response Parameters on Board / A. P. Nyrkov, M. V. Ka-rdakova, S. V. Kolesnichenko, [et al.] // Proceedings of the 2020 IEEE Conference of Russian Young Researchers in Electrical and Electronic Engineering (ElConRus), (Saint Petersburg, Moscow, Russia, 27-30 January 2020). — Institute of Electrical and Electronics Engineers, 2020. — Pp. 434-437. DOI: 10.1109/EIConRus49466.2020.9038955.
17. The Use of Fuzzy Control Methods for Evaluation of Complex Systems on the Example of Maritime Fleet Equipment / A. P. Nyrkov, A. A. Zhilenkov, S. S. Sokolov, S. G. Chernyi // Vibroengineering PROCEDIA. 2016. Vol. 8. Pp. 6-10.
18. Mathematical and Information Maintenance of Bio-metric Systems / Z. V. Boriev, S. S. Sokolov, A. P. Nyrkov, A. A. Nekrasova // Proceedings of the International Conference on Mechanical Engineering, Automation and Control Systems 2015 (MEACS 2015), (Tomsk, Russia, 01-04 December 2015). IOP Conference Series: Materials Science and Engineering. 2016. Vol. 124. Art. No. 012046. 4 p.
DOI: 10.1088/1757-899X/124/1/012046.
19. Svetlov, V. A. Development of the Algorithm of Adaptive Construction of Hierarchical Neural Network Classifiers / V. A. Svetlov, S. A. Dolenko // Optical Memory and Neural Networks (Information Optics). 2017. Vol. 26, No. 1. Pp. 40-46. DOI: 10.3103/S1060992X17010076.
20. Modeling of the Multilayer Perceptrons for Image Recognition of the Steel Microstructures / V. A. Yemelyanov, N. Yu. Yemelyanova, E. V. Shved, [et al.] // Proceedings of the 2020 IEEE Conference of Russian Young Researchers in Electrical and Electronic Engineering (ElConRus), (Saint Petersburg, Moscow, Russia, 27-30 January 2020). — Institute of Electrical and Electronics Engineers, 2020. — Pp. 952-955. DOI: 10.1109/EIConRus49466.2020.9038971.
21. Dynamics Models of Synchronized Piecewise Linear Discrete Chaotic Systems of High Order / S. S. Sokolov, A. A. Zhilenkov, S. G. Chernyi, [et al.] // Symmetry. 2019. Vol. 11, Is. 2. Art No. 236. 12 p. DOI: 10.3390/sym11020236.
22. Kotenko, I. V. Analytical Modeling and Assessment of Cyber Resilience on the Base of Stochastic Networks Conversion / I. V. Kotenko, I. B. Saenko, O. S. Lauta // Proceedings of the 10th international workshop on Resilient Networks Design and Modeling (RNDM 2018), (Longyearbyen, Norway, 27-29 August 2018). — Institute of Electrical and Electronics Engineers, 2018. — Pp. 1-8.
DOI: 10.1109/RNDM.2018.8489830.
23. Providing Security for Virtual Infrastructures Based on an Integrated Mechanism / S. S. Sokolov, O. M. Alimov, Yu. A. Trubacheva, [et al.] // Proceedings of the 2018 IEEE Conference of Russian Young Researchers in Electrical and Electronic Engineering (ElConRus), (Moscow, Saint Petersburg, Russia, 29 January-01 February 2018). — Institute of Electrical and Electronics Engineers, 2018. — Pp. 132-135. DOI: 10.1109/EIConRus.2018.8317047.
24. Saenko, I. B. Detection of Traffic Anomalies in MultiService Networks Based on a Fuzzy Logical Inference / I. B. Saenko, S. A. Ageev, I. V. Kotenko // Intelligent Distributed Computing X (IDC2016): Proceedings of the 10th International Symposium on Intelligent Distributed Computing (Paris, France, 10-12 October 2016) / C. Badica, [et al.] (eds.). — Cham: Springer Nature, 2017. — Pp. 79-88. — (Studies in Computational Intelligence. Vol. 678).
DOI: 10.1007/978-3-319-48829-5 8.
