Classification of hybrid systems. Subject and purpose of development of hybrid intelligent systems

The entire development and operation cycle of any complex system is iterative (Fig. 3.12). Performing any iteration as shown in Fig. 3.12 is carried out using models of a complex system. The most advanced and powerful tool for constructing appropriate models for the systems under consideration is simulation modeling. it provides a deep representation of the modeled object, makes it possible to analyze processes at any time interval, allows you to take into account random and uncertain factors, and evaluate both technical and economic indicators of the system’s functioning.

Rice. 3.12. Development cycle of a complex system

A complex system is a system with evolution and is characterized by a large number heterogeneous subsystems with a high degree of uncertainty. Consequently, solving problems of analysis, control and others in such systems cannot be carried out using any single approach for all subsystems.

Decision making usually uses a complex combination of mathematical, statistical, computational, heuristic, experimental and engineering methods (most often expert systems). The integrated use of these methods and tools provides the user with support when making decisions. In this case, the priority of the problem being solved takes place over the methods used.

Existence similar situation, when it is necessary to use simulation and different decision-making methods together, has led to the emergence of so-called hybrid systems. By hybrid system we mean a system consisting of several systems various types, the functioning of which is united by a single goal (Fig. 3.13).

Rice. 3.13. The simplest hybrid system

The simplest hybrid system is a system that combines a simulation model and an optimization unit. The optimization block implements one of the algorithms search engine optimization(for example, the simplest genetic algorithm - PGA), and simulation model serves to calculate the values ​​of the optimization criterion (suitability function) for the selected solution options.

Running a simulation model provides, at best, obtaining results at one point in the solution search space. Therefore, it is necessary to implement a series of experiments on a simulation model in a large search area, the focus of which is ensured in traditional modeling systems by a specialist developer.

The use of genetic algorithms to solve optimization problems in the analysis, control or synthesis of truly complex systems is only possible if there is a way to determine the fitness function of an individual with sufficiently good accuracy. That is, it is necessary to be able to develop models of complex systems with a high degree of adequacy to objects and processes of the real world.

Let's consider hybrid systems that use a genetic algorithm and simulation together to solve problems of various types (Fig. 3.14). This, first of all, relates to the tasks of organizational management, real-time decision-making, evaluation of management strategies, and forecasting.

Rice. 3.14. The simplest hybrid system with genetic algorithm and simulation model

The purpose of the hybrid system optimization block is to improve the solution by selecting the values ​​of the controlled variables. PGA is used for these purposes. The genetic algorithm can be implemented on any universal language, for example, C++, Pascal, etc. However, a hybrid system built on a single software, for many reasons, is preferable to a system that combines blocks written in different software.

Existing methods and simulation languages ​​are often ineffective due to their low flexibility and complexity in modeling decision-making and control systems, especially if the control system includes a human operator making decisions. use of those available on the market software products intelligent systems Simulation modeling removes some of these difficulties and provides new opportunities for using simulation in hybrid systems to solve applied system problems.

The hybrid system implements the functions of not only an intelligent interface, but also an intelligent computer. The composition of a typical hybrid system, including the indicated components, is shown in Fig. 3.15.

Calculation block Simple genetic algorithm

optimality criterion

Rice. 3.15. Structure of a typical hybrid circuit

In this scheme, the simulation model serves to draw up a plan and uses a set of heuristic rules to determine the priority of a particular order included in the work plan.

The optimization block ensures the selection of priority rules for drawing up work plans with the best performance. Choice required best rules For current situation, as well as choice optimal values their parameters.

The purpose of the expert system as part of a hybrid system is to improve the PGA performance, first of all, to increase the convergence of the optimization process by including into the process some ideas (knowledge) of a human operator about the prospects of a particular search strategy. In this case, the expert system performs the function of a “selector,” purposefully changing the PGA parameters to reduce computation time.

The expert system carries out a directed selection of such PGA parameters as: population size, probabilities of crossing and mutation. In addition, she applies some rules to preserve individuals with high value fitness functions from generation to generation during reproduction, etc.

An expert system thus represents a field of combinations of knowledge about genetic algorithms, computational mathematics, artificial intelligence and expert knowledge. The application domain for the expert system is not well defined and the search space is poorly structured and therefore the expert system works alongside the PGA based on current data regarding the population and the current state of the simulator.

It looks like the days of the classic internal combustion engine are numbered. And this is due to the need to comply with environmental standards.

The Euro 6 standard, which came into force on September 1, 2015, has tightened the requirements for diesel engines regarding the content of nitrogen oxides and residual hydrocarbons in the exhaust. Compliance with established toxicity standards is ensured catalytic reduction system , exhaust gas recovery system, improved particulate filter. From September 1, 2017, certification of new cars will be carried out according to a more stringent global driving cycle and availability listed systems on diesel engines will not be enough to meet established toxicity standards.

In accordance with the decision of the European Parliament Commission, starting from 2021, a carbon dioxide emission standard of 95 g/km will be introduced for all manufactured cars instead of the current 130 g/km. This corresponds to an average fuel consumption of 4.06 l/100 km for gasoline engines and 3.62 l/100 km for diesel engines. This level of fuel consumption is unattainable for classic internal combustion engines.

Therefore, hybrid vehicle drive designs that fully meet environmental requirements are coming to the fore. But modern hybrids are very expensive. A compromise solution is the so-called low voltage hybrid system. The system is based on 48 volt electrical network, which is planned as an addition to the main 12-volt on-board network.

Main functions of the low voltage hybrid system:

The use of a low-voltage hybrid system in vehicles can reduce carbon dioxide emissions by 10-15%. The low-voltage hybrid with a diesel engine reduces nitrogen oxide emissions by 20%, and together with the catalytic reduction system - by 80%. Fuel savings can reach from 13 to 21%.

Today, several leading manufacturers are developing low-voltage hybrid systems, including Bosch, Continental, Delphi, Ricardo, Valeo. Currently, most automakers have decided to implement it, and Audi, Honda, Ford, Kia, Renault, and Volkswagen are already installing a low-voltage hybrid system on their cars. It is predicted that by 2020, 25% of new cars will have a hybrid drive, and half of them will use a 48-volt electrical system.

A typical low-voltage hybrid system design consists of the following main elements included in the 48-volt electrical network: starter-generator, inverter, converter direct current and a 48-volt battery.

The starter-generator is the main structural element systems. It operates in two modes - as a generator and as a starter (more precisely, as Electrical engine). In generator mode it is created Electric Energy, which allows you to completely abandon the traditional 12-volt generator. The starter mode is used to create additional torque when the vehicle is moving. Synchronous and asynchronous AC electric machines are used as starter generators.

Specific view electric machine determined by the design diagram of the hybrid installation. There are several design schemes for a low-voltage hybrid power plant:

1. The starter-generator is connected to the internal combustion engine by a belt drive.
2. The starter-generator is installed on the crankshaft of an internal combustion engine.
3. The starter-generator is located in the gearbox.

The simplest and, therefore, by far the most common scheme for integrating a starter-generator is to connect it to an internal combustion engine using a belt drive.

The operation of the starter-generator is ensured by a bidirectional inverter. It converts the battery's DC current into three-phase alternating current. In energy recovery, the inverter converts alternating current into direct current to charge the battery. The design of the 48-volt inverter is similar to the high-voltage inverter used in full hybrids and electric vehicles.

A DC/DC converter is used to transfer power between 48-volt and 12-volt electrical systems. For 12 volt electrical network it replaces the generator.

Each network has its own battery: a 48-volt lithium-ion battery and a 12-volt lead-acid battery. 48 volt accumulator battery serves to power powerful devices: air conditioning compressor, water pump, electromechanical rotary shock absorbers, active suspension, etc. The 12-volt battery provides energy lighting system, security systems, infotainment system.

Under hybrid intelligent system(GIS) is commonly understood as a system in which more than one method of simulating human intellectual activity is used to solve a problem. Thus, GiIS is a collection of:

• analytical models
• expert systems
• artificial neural networks
• fuzzy systems
• genetic algorithms
• statistical simulation models

The interdisciplinary direction “hybrid intelligent systems” brings together scientists and specialists who study the applicability of not one, but several methods, usually from various classes, to solving management and design problems.

Encyclopedic YouTube

1 / 2

Alexey Paevsky: What's new in medicine

15x4 Comic-Con: 15 minutes about EVE Online ships

Subtitles

History of the term

The term “intelligent hybrid systems” appeared in 1992. The authors put into it the meaning of hybrids of intelligent methods, such as expert systems, neural networks and genetic algorithms. Expert systems were symbolic, and artificial neural networks And genetic algorithms- adaptive methods of artificial intelligence. However, basically new term concerned a rather narrow area of ​​integration - expert systems and neural networks. Below are several interpretations of this area of ​​integration by other authors:

1. " Hybrid approach" suggests that only a synergistic combination of neural and symbolic models achieves the full range of cognitive and computational capabilities (abilities).

2. The term “hybrid” is understood as a system consisting of two or more integrated subsystems, each of which may have various languages representations and inference methods. Subsystems are combined together semantically and in action, each with each.

3. Scientists of the Center Artificial Intelligence Cranfield University (England) define a “hybrid integrated system” as a system that uses more than one computer technology. Moreover, technologies cover such areas as knowledge-based systems, connectionist models and databases. Technology integration makes it possible to use the individual power of technology to solve specific parts of a problem. The choice of technologies introduced into a hybrid system depends on the characteristics of the problem being solved.

4. Specialists from the University of Sanderland (England), members of the HIS (Hybrid Intelligent Systems) group, define “hybrid Information Systems"like big ones, complex systems, which “seamlessly” (seamlessly) integrate knowledge and traditional processing. They can provide the ability to store, search and manipulate data, knowledge and traditional technologies. Hybrid information systems will be significantly more powerful than extrapolations of existing systems concepts.

Subject and purpose of development of hybrid intelligent systems

The scientific field of GiIS includes the study of autonomous methods to determine their advantages and disadvantages, integration relations that largely determine the composition, architecture and processes of exchange and processing of information in hybrids, identification of tasks corresponding to hybrid systems, development of protocols for communication between components and multiprocessor architectures.

The goals of GIS research include the creation of methods for increasing the efficiency, expressive power, and inference power of intelligent systems, preferably more complete, developed with less development effort than applications using stand-alone methods. From a fundamental perspective, GIS can help to understand cognitive mechanisms and patterns.

Classification of hybrid intelligent systems

Based on an analytical review of existing GIS classifications, it is proposed to distinguish the following five GIS development strategies: autonomous, transformational, loosely coupled, tightly coupled and fully integrated models:

• Autonomous GIS application models contain independent software components, implementing information processing on models using methods from limited number classes. Despite the obvious degeneracy of knowledge integration in this case, the development of autonomous models is relevant and can have several goals. Such models are a way of comparing the capabilities of solving a problem by two or more different methods. A new autonomous model for solving a solved problem verifies the already created application and leads to adequate models. Stand-alone models can be used for quick creation an initial prototype, followed by development of more time-consuming applications. Autonomous models also have a significant drawback - none of them can help the other in the situation of updating information - all must be modified at the same time.

• Transformational GIS are similar to autonomous ones, since final result development - an independent model that does not interact with other parts. The main difference is that such a model begins to work as a system using one autonomous method, and ends up as a system using a different method. Transformational models provide several advantages: speed of creation and lower costs, since a single model is used, and the final method the best way adapts results to the environment. There are also problems: automatic conversion one model to another; a significant modification of the model, comparable in scope to a new development.

• Tightly connected GIS have low communication costs and more high performance compared to loosely coupled models. However, these GIS also have three fundamental limitations: 1) the complexity of development and support increases as a consequence of the external data interface; 2) strong coupling suffers from excessive accumulation of data and 3) verification of adequacy is difficult. Due to the fact that their composition and structure largely depend on the problem being solved, the considered weakly and strongly coupled GIS are also called functional GIS.

• Fully integrated GIS shares general structures data and knowledge representation, and the relationship between the components is achieved through the dual nature of the structures. This is a rapidly developing class of hybrids in world practice, where we can highlight the development of conceptual neural networks based on knowledge, connectionist expert systems in which elements interact quickly and simply, and general information for an independent solution of the problem, it is instantly available to both components. Another option for complete integration is fuzzy neural networks - a hybrid that is similar in structure to a neural network and implements both neural and fuzzy calculations. The advantages of full integration are reliability, increased processing speed, adaptation, generalization, noise reduction, argumentation and logical deduction, something that cannot be found in total in any class of parent methods.

Results

As part of the study of methodologies for creating GIS in 2001, a problem-structural methodology and technology for developing GIS were proposed, allowing for the synthesis of GIS to solve complex ones (consisting of many subtasks that require the use of various methods imitation of human intellectual activity) tasks as a system of methods for solving subtasks of a complex task. Later in 2007, a problem-tool methodology for developing GIS was proposed as a generalization of the problem-structural methodology in the case of the absence of relevant methods for solving subtasks of a complex problem.

Based on the proposed methodologies and technologies, GIS has been developed for practical application V various areas: shift-daily planning in a seaport, planning in a bioproduction system, design of automation of sea transport vessels, solving complex transport and logistics problems, medium-term planning at a manufacturing enterprise with a small-scale production nature and others. Detailed description listed GiIS and their results practical use can be found in relevant sources.

see also

List of used literature

1. Kolesnikov A.V., Soldatov S.A. Theoretical foundations of solving the complex problem of operational production planning taking into coordination. // Bulletin of the Russian State University. Immanuel Kant. Vol. 10: Ser. Physical and mathematical sciences. – Kaliningrad: Publishing house. RSU named after. I. Kant, 2009. – P. 82-98.

1. Kolesnikov A.V. Hybrid intelligent systems: Theory and technology of development / Ed. A.M. Yashina. - St. Petersburg. : Publishing house of St. Petersburg State Technical University, 2001. - 711 p. - ISBN 5-7422-0187-7.
2. Gavrilov A.V. Hybrid intelligent systems. - Novosibirsk: NSTU Publishing House, 2003. - 168 p.
3. Yarushkina N.G. Fundamentals of the theory of fuzzy and hybrid systems. - M.: Finance and Statistics, 2004. - 320 p.
4. Kolesnikov A.V. , Kirikov I.A. Methodology and technology of solution complex tasks methods of functional hybrid intelligent systems. - M.: IPI RAS, 2007. - 387 p. - Kolesnikov A.V. , Kirikov I.A. , Listopad S.V. , Rumovskaya S.B. ,Domanitsky A.A. Solving complex traveling salesman problems using functional hybrid intelligent systems methods / Ed. A.V. Kolesnikova. - M.: IPI RAS, 2011. - 295 p. - ISBN 978-5-902030.
5. Klachek P.M. , Koryagin S.I. , Kolesnikov A.V. , Minkova E.S. Hybrid adaptive intelligent systems. Part 1: Theory and technology of development: monograph. - Kaliningrad: Publishing house of the IKBFU. I. Kant, 2011. - 374 p. - ISBN 978-5-9971-0140-4.
6. Kolesnikov A.V. , Soldatov S.A. Theoretical basis solving the complex problem of operational and production planning taking into account coordination // Vestnik Rossiiskogo state university them. Immanuel Kant. - Kaliningrad: Publishing house. RSU named after I. Kant, 2009. - Vol. 10: Ser. Physical and mathematical sciences. - pp. 82-98.
7. Medsker L.R. Hybrid Intelligent Systems. - Boston: Kluwer Academic Publishers, 1995. - 298 p.
8. Wermter S., Sun R. Hybrid Neural Systems. - Heidelberg, Germany: Springer-Verlag, 2000.
9. Negnevitsky M. Artificial Intelligence. A guide to intelligent systems. - Harlow, England: Addison-Wesley, 2005.
10. Castillo O., Mellin P. Hybrid Intelligent Systems. - Springer-Verlag, 2006.
11. Jain L.C. , Martin N.M. Fusion of Neural Networks, Fuzzy Systems and Genetic Algorithms: Industrial Applications. - CRC Press, CRC Press LLC, 1998.

Hybrid intelligent system

Under hybrid intelligent system It is customary to understand a system in which more than one method of simulating human intellectual activity is used to solve a problem. Thus, GIS is a collection of:

• analytical models
• expert systems
• artificial neural networks
• fuzzy systems
• genetic algorithms
• statistical simulation models

The interdisciplinary direction “hybrid intelligent systems” brings together scientists and specialists who study the applicability of not one, but several methods, usually from different classes, to solving control and design problems.

History of the term

The term “intelligent hybrid systems” appeared in 1992. The authors put into it the meaning of hybrids of intelligent methods, such as expert systems, neural networks and genetic algorithms. Expert systems were symbolic, and artificial neural networks and genetic algorithms were adaptive methods of artificial intelligence. However, basically, the new term concerned a rather narrow area of ​​integration - expert systems and neural networks. Below are several interpretations of this area of ​​integration by other authors.

Prerequisites

1. The “Hybrid Approach” assumes that only a synergistic combination of neural and symbolic models achieves the full range of cognitive and computational capabilities.

2. The term "hybrid" is understood as a system consisting of two or more integrated subsystems, each of which may have different presentation languages ​​and output methods. Subsystems are combined together semantically and in action, each with each.

3. Scientists at the Center for Artificial Intelligence at Cranfield University (England) define a “hybrid integrated system” as a system that uses more than one computer technology. Moreover, technologies cover such areas as knowledge-based systems, connectionist models and databases. Technology integration makes it possible to use the individual power of technology to solve specific parts of a problem. The choice of technologies introduced into a hybrid system depends on the characteristics of the problem being solved.

4. Specialists from the University of Sanderland (England), members of the HIS (Hybrid Intelligent Systems) group, define “hybrid information systems” as large, complex systems that “seamlessly” (seamlessly) integrate knowledge and traditional processing. They can provide the ability to store, search and manipulate data, knowledge and traditional technologies. Hybrid information systems will be significantly more powerful than extrapolations of existing systems concepts.

see also

Literature

• Gavrilov A.V. Hybrid intelligent systems. – Novosibirsk: NSTU Publishing House, 2003. – 168 p., ill.
• Kirikov I. A. Methodology and technology for solving complex problems using the methods of functional hybrid intelligent systems. - M.: IPI RAS, 2007. - 387 p., ill. - ISBN 978-5-902030-55-3
• Larry R.Medsker. Hybrid Intelligent Systems. 1995.
• Stefan Wermter, Ron Sun, Hybrid Neural Systems. Springer-Verlag, Heidelberg, Germany. 2000.
• Negnevitsky M. Artificial Intelligence. A guide to intelligent systems. Addison-Wesley, 2005.
• Castillo, P. Mellin, Hybrid Intelligent Systems, Springer-Verlag. 2006.
• Lakhmi C. Jain; N.M. Martin Fusion of Neural Networks, Fuzzy Systems and Genetic Algorithms: Industrial Applications. - CRC Press, CRC Press LLC, 1998

Links

• International Conference on Hybrid Intelligent Systems
• International Journal of Hybrid Intelligent Systems
• Website with information about hybrid intelligent systems

Wikimedia Foundation. 2010.

See what a “Hybrid intelligent system” is in other dictionaries:

Not to be confused with analog digital computing system. Hybrid computing system system with heterogeneous hardware computing structure. Combination of any computing devices or blocks, such as calculations with using the CPU and... ...Wikipedia

The request for "Intelligent Dialog System" is redirected here. A separate article is needed on this topic. Intelligent system (IS, English... Wikipedia Wikipedia

This article should be Wikified. Please format it according to the article formatting rules. The JSM method is a method for automatically generating hypotheses. Formalizes a scheme for a plausible and reliable conclusion, n ... Wikipedia

- (AI) (eng. Artificial intelligence, AI) is the science and development of intelligent machines and systems, especially intelligent computer programs aimed at understanding human intelligence. In this case, the methods used are not required... ... Wikipedia

Artificial intelligence (AI) is the science and development of intelligent machines and systems, especially intelligent computer programs, aimed at understanding human intelligence. At the same time... ... Wikipedia