Lukina Yulia

Faculty of Computer Science and Technology
Department of Applied Mathematics and Computer Science
Speciality "Software engineering"

Scientific adviser: Candidate of Technological Science Oleg Fedyaev

Abstract on theme "Agent-oriented software models of human behavior in socio-economic environment"

Aims and objectives

The aim is to study multi-agent systems, possible applications of agent-basedapproach to solve the problem, and implementation of software using the agent system MadKit.
Research objectives:

  • Review of methodologies and tools for analyzing and designing multi-agent systems.

  • Building agent-based models.

  • The choice of the architecture of software agents.

  • The choice of method of agent interaction.

  • Analysis tool environment MadKit.

  • Software implementation of multi-agent system

  • Analysis of simulation results.

Relevance of the topic

Multi-agent systems (MAS) – one of the promising new areas of artificial intelligence, which was formed on the basis of research results in distributed computer systems, networking technologies to solve problems and parallel computing.
The urgency of the MAS is currently subject to the following main reasons:

  • The complexity of modern systems and organizations, which reaches such a level that centralized management of them becomes ineffective because of the huge flows of information;

  • Tasks or developed systems often heterogeneous and distributed in space and in functional terms, as no one person can not build a modern complex system alone; 

  • The evolution of software occurs in the direction of its development, based on autonomous, individualized, interactive modules;

  • Distribution of the various networks generates a distributed view of the world.

In general, multi-agent technology provides the following important benefits:

  • New generation of innovative products and carry a qualitatively new features: a well-known or new complex problems are solved by a new productive and efficient manner;

  • Characterized by flexibility: they work dynamically respond to changes and continuously improve the solutions in real time;

  • Intellectual: agents do not just follow a given process, but also analyze the situation and looking for a way to solve the problem that is guaranteed to find the best possible solutions;

  • Take into account with varying degrees of importance, even the smallest factors necessary for decision-making: decisions are personalized, individualized and meet the requirements of all participants;

  • The high-performance;

  • Characterized by a high rate of reaction to events and quickly find a solution to the problem;

  • Allow you to adjust the results of the system;

  • Are able to learn.

The main disadvantage of MAS is the unpredictability of the behavior of the complete system [2], based on its constituent components, as there is no "common" security management. As a result, the agent can act as an intruder and use a system of "fraudulently".
Agent-oriented environment can serve as an effective means of studying, researching and solving complex problems in a wide range of subject areas.

Scientific novelty

Modeling the behavior of persons isn't avaible using mathematical tools, because it is impossible to describe human behavior limited set of mathematical functions.Therefore, agent-oriented approach is ideal for modeling human behavior, because intelligent agent initially has the basic knowledge, based on which he behaves in certain ways. Over time, intelligent agent "gets experience", change or expand the knowledge and, consequently, changed behavior. I developed the system used a multi-agent environment MadKit, which accounts fordifference from all other similar systems.

Expected practical results

Practical results are as follows:

  • Development of a new scientific field multi-agent systems.

  • Improve the quality of modeling socio-economic processes through the use of intelligent agents.

  • Analysis of simulation results, the estimate approximate model of reality.

A review of research and development

Multiagent approach is used to model the economic system in [4]. In [4] identified four types of non-stationarity of the environment and an algorithm for solving the three of them. The algorithm is based on the use of multi-agent approach, which allows you to organize distributed computing, which will reduce the execution time given algorithm can be implemented without the use of multi-agent system. However, for sufficiently large amount of computation is large enough. Formalize the prediction in a nonstationary fourth kind is difficult, because this type of nonstationarity requires revolutionary changes in the environment, and as a consequence, the unsuitability of the accumulated data for the projections. In such a situation to solve the problem of forecasting, appropriate use of experts or to construct nonlinear nonparametric model such as multi-agent, algorithms, behavior of agents which represent the direction of future research.
Using multi-agent approach in telemedicine is presented in [5]. The article presents the clinical experience of teleconsultation in Uzbekistan, discussed a new concept of telemedicine intellectualization, implemented based on Grid technology and multi-agent systems. Grids provide the infrastructure, protocols and middleware software, by which one can discover, aggregate and develop resources. Agent technology creates the autonomous problem solvers that can purposefully and convincingly operate in a dynamically changing environment of Grid. The combination of these technologies creates a new perspective to the development of theoretical and applied research in telemedicine.
Mulagentny approach is also used to simulate cellular automate. Analytical simulation of regular and statistically homogeneous large networks (including networks of queues), undertaken in order to identify the asymptotic behavior of their structural properties and developing the scheme Cellular Automate – grilles – large networks implemented in its experimental of a modeling environment RepastS. This environment provides advanced multi-agent simulation tools, powerful, interactive graphical interface, the methods of creating a variety of grids (net spaces of agents) and has high productivity [6]. Preliminary experiments demonstrate the feasibility studies and programs outlined in its theoretical part: as the size of a regular or statistically homogeneous structures are found stable asymptotic laws, identifies the key parameters are defined and their critical values, which allow to effectively solve the problem of synthesizing large networks.
The need to take a large number of administrative decisions of the municipal administration of Education (MAE) to develop and upgrade the urban infrastructure, makes actual use of automated decision support systems (DSS). DSS based on a simulation model of MAE can be a tool for comprehensive analysis and forecast of the situation in the city, allowing you to assess the possible risks of the implementation of these projects, their influence on each other and the quality of life of citizens [7]. Appropriateness of the unit multi-agent simulation to predict the consequences of finding effective management decisions in the MAE is obvious. Interaction between subsets of agents to each other in certain circumstances, taking into account the decisions taken by the Administration, gives a comprehensive picture of the situation in the city. The introduction of a fully functional DSS will reduce the time it takes to make a decision, the financial cost of "manual" calculations and data collection, and allow interactive "play" a variety of options for situations that will enhance the validity of the strategic decision-making senior and middle management of MAE.
Multi-agent simulation is used to study the mechanisms of protecting the information on the Internet in article [8]. This article describes a proposed approach to multi-agent simulation of information security. Approach is presented as an example of an antagonistic confrontation between two teams of agents: agents of implementation attacks and protection agents of this class of attacks. The features of the approach described by developed simulation environment and presents some of the scenarios modeling. The experiments showed the possibility of using the proposed approach for modeling the mechanisms of protection and for the analysis of designed networks. They also demonstrated that the use of co-operation of several teams and combined use of various adaptive protection mechanisms leads to substantial improvements in its effectiveness. Directions for further work related to the investigation of mechanisms of protection against different types of attacks, as well as the improvement of the modeling system.
For the modeling of market relations is also used multi-agent system in [9]. For industry, this approach will improve the competitiveness of the projects by improving the graphics, effective use of resources, closer interaction among the subcontractors. Method is the distributed coordination framework for developing multi-agent planning and management system that will help sub-contractors to make the timing of actions, identify and analyze their own resource constraints in this graphic, predict the result of replanning actions to coordinate different perspectives of planning, working towards better solutions.
Currently, many businesses and organizations there is a need for rapid analysis of the market situation and decision in the shortest possible time. To minimize the risks associated with inaccurate and unjustified decisions, developed a decision support system (DSS). The main function of DSS is to generate different alternatives for making a certain situation that has arisen in a particular process of converting resources (PCR). Among the methods of modeling the most problematic area is full of PCR corresponds to the multi-agent simulation [10]. As a result of research developed software simulation and decision support systems, allowing to develop effective management solutions for enterprises and organizations of various types, to create models of the processes of transformation resources with the possibility of developing scenarios of the behavior of agents and coalitions, conduct simulation experiments using the mechanism of plans of action of agents and coalitions.
Multi-agent simulation of firms competition in the market, the struggle to maximize profits and market share is described in [11]. The above results reveal a multi-agent simulation and explain the mechanism of pricing strategies of active elements in the marketing environment and allow to predict the processes of stabilization of the market in various economic and social disturbances with the choice of the best marketing strategies according to supply and demand in the current environment.
Development of virtual cathedra is an urgent task of the modern educational process, as it allows to transfer the load on teachers at the system more flexible to interact with students, use an individual approach. The development of MAS training students in the discipline is given in [12, 13]; remote MAS in [14].
There is also the prospect of using multiagent approach in the creation of the Grid. The main task of the Grid - coordinating collections of resources. Using agent-based systems for planning tasks in the Grid will provide an opportunity to unleash the two major problems – scalability and adaptability. Application of Grid systems can significantly increase the speed and quality of computing, especially in the case of loosely connected heterogeneous computing systems. In [17] described the architecture of the modern grid system based on software agents.
Also actual design multi-agent systems for teams of robots. In [18] considered group (team) algorithms for strategic and tactical levels of management team of robots. The analysis of the schemes of robot soccer competition shows that, despite the different designs of such robots, and technical decisions taken in their construction, in the schemes of robot competitions have much in common in the higher, strategic levels of management players that can pose the problem of mining by unified modeling tools. As such a system in this paper the design of the control algorithm used a system of "virtual football".

The theory of agent-oriented systems
Intelligent agent

Intelligent called agent that can act autonomously and flexibly to achieve its goals, with a flexibility to understand:

  • reactivity or efficacy: intelligent agents are able to perceive their environment and react to changes that occur in the environment, to achieve their goals;

  • pro-activity or dedication: intelligent agents are able to demonstrate goal-directed behavior by taking initiative in achieving their goals;

  • the ability for social ability or a collectivity: intelligent agents are able to interact with each other (and perhaps with humans) in order to achieve their goals.

Generalized functional structure of an agent consists of 5 blocks (Fig. 1):
Figure 1 – Generalized functional structure of the agent (S – sensory system, E – estimates block, D – decisions block, A – action system, C – communicate block)

  • S (sense) – sensory system, is responsible for receiving information about the state of the environment, for example, in the form of parameters that are measured by sensors, sensory (temperature, pressure, radioactivity), or in the form of images obtained using a video camera.

  • C (communicate) – unit of information exchange with other agents, provides information exchange certain content and format of the neighboring agents.

  • E (estimate) – unit of evaluation; forms the signal win or loss on the basis of information about the current state of the environment and information from the unit of information exchange.

  • D (decide) – block the decision-making is responsible for choosing the next action based on information about the success of previous actions (example: an automaton with linear tactics, which ensures convergence to the winning solution in a stationary random environment).

  • A (actuate) – action system, ensures the execution (implementation) of selected actions (decisions) (for example, implements the transfer agent in the space in the chosen direction).

Multiagent systems

Multi-agent systems (MAS) – the union of separate intelligence systems.

You can give a formalized definition of multi-agent systems:
MAS = (A, E, R, ORG, ACT, COM, EV), (1)
where MAS – multi-agent system, A – ​​set of agents, E – the set of environments that are in certain respects, R and interacting with each other, forming some organization ORG, having a set of individual and consistent action ACT (the strategy of behavior and actions), including possible communication actions COM and the possibility of the evolution of EV.
To make the MAS is ready for industrial application, a nonprofit association called FIPA, proposed a series of norms and standards that developers of multi-agent systems must comply to MAS have been compatible with other systems:

  • An agent can communicate with other agent

  • The agent provides a range of services that are available to any agent in the system.

  • Each agent is obliged to limit its availability from other agents.

  • Each agent is obliged to determine its attitude, contracts, etc. with other agents. Thus, the agent "knows"just a set of agents with whom he can communicate.

  • Each agent has with his name my way (the concept of ID Agent). Therefore, agents are supposed to stand alone, and there are no restrictions on the way to interact.

MadKit

As a tool environment development system was chosen MadKit. MadKit is a set of packages, classes, Java, which implements the core agent, various libraries of messages and agents. It also includes a graphical development environment and the standard model of the agent. Architecture MadKit based on a very small kernel. Basic services such as: distributed messaging, migration and control – implemented by agents platform for maximum flexibility. Component interface model allows the interfaces to various agents and manages a global GUI.
Architecture MADKIT based on the AGR (Agent / Group / Role) model, known as AALAADIN (Fig. 2). AGR model is based on three primitive concepts: agent, group and role, which are structurally interconnected and can not be defined by other primitives.
модель АГР
Figure 2 – Model AALAADIN
Agent is an active object of sending messages, which plays a role within the group. An agent can play several roles, and may be a member of several groups. One of the most important characteristics of the model AGR is that there are no restrictions imposed on the internal architecture of the agent's behavior and its possibilities. It is important to make the model as universal as possible.
Group is a number of agents that share some common characteristics. The agent can be a member of several groups at one and the same time. The important point groups AALAADIN is that they can freely overlap. A group can be established by any agent.
Role is an abstract representation of the functional position of the agent in the group. Agent must play a role in the group; agent can play several roles. Roles are local to the group, and the agent must request a role for its implementation. Several agents may play one role. Thus, the role is an abstract concept that can be played one or many agents.

Model agent in MadKit

The structure of the agent in MadKit shown in Fig. 3.
структура модели агента 
Figure 3 – The structure of the model agent MadKit
Agent in MadKit consists of 4 mandatory sections [24]:

  • Activate section containing a code to be executed immediately after the creation of an agent.

  • Live section, which contains the basic code that describes the behavior of the agent. Normally, this section contains an infinite loop.

  • End section, contains some code that is executed when the agent is destroyed.

  • Graphical section (initGUI section), containing a description of component Java, which should be used as a graphical user interface agent, and is intended to replace the GUI by default.

MadKit does not impose on the architecture of agents is no restriction for to maximize the universality of applications.

Interaction of agents in MadKit

Interaction of agents in MadKit by using asynchronous messaging. An agent can send a message to another agent, defined by its address or by using a broadcast message that is sent to agents playing this role in a particular group.
MadKit provides several types of predefined messages, such as StringMessage, XMLMesssage and ActMessage. With the latest type of messages you can define the following sub-messages: ACLMessages and KQMLMessages [24]. It is also possible to define your own message class, which will inherit from the default message.
Each agent has its own "mailbox" in which messages are delivered and who should check the agent to receive the message.

An example of an agent based model of the spread of the virus

With the help of MadKit established agent model the spread of viruses. Simulated environment is a set of objects (supposably, humans), a certain number who are infected with the virus, and the rest aren't. All the objects move in a certain way in any direction. Infection occurs through direct collision of healthy and infected objects, or may occur in a certain radius from the infected object.
Several states modeled the spread of viruses is shown in Fig. 4. In this example, the healthier items are marked in green, infected are red.
моделирование процесса распространения вируса
Figure 4 – Modelling of process of distribution of a virus by means of tool MadKit environment (Animation – the permission: 196 x 214 px; volume: 36.6 kb; shots: 6; a delay between shots: 0.5с; quantity of repetitions: 6)
Parallel visual modeling step plotted graphic depending on the number of infected objects from time.
With window interface can set the initial parameters of the simulation.
This model can be used to study the following processes:

  • flu epidemic in crowded places;

  • spreading rumors on social networking sites;

  • dissemination of information in working groups, student groups, a group of relatives, a group of acquaintances, etc.

Conclusions

Instrumental environment MadKit is a universal multi-agent platform. This tool is based on an organizational model AGR. Architecture MadKit based on a microkernel [24]. Basic services such as: distributed messaging, migration or control are implemented by agents platform for maximum flexibility. Component interface model allows the interfaces to various agents.
There are examples of successful use MadKit in projects related to a wide range of applications of modeling hybrid architecture for control of underwater robots to evaluate the social networks or research of multiagent control on the production line.
The methodology used in MadKit, is sufficient easy to use.
The big advantage of this tool environment is the fact that it imposes no restrictions on the architecture of agents to achieve maximum flexibility of applications. For example, now widely recognized as the BDI-architecture, and after a while, it may be used by another new architecture of intelligent agent. In this case, the developers of agent-based systems will not need to learn another tool environment for the new architecture.
Interaction between agents by using different types of messages, also have the opportunity to create their own message types. A large variety of types of messages can be attributed to the positive characteristics of the environment. Agents can interact both online and on the same computer from the same or different sessions of the simulation.
MadKit system is a convenient tool for the development of MAS, because it has lots of visual elements, which simplify the work. For example, in MadKit have a tree relationship of agents, which contains information about which group includes agents, role play, etc., the message table that contains information about the agent-sender, recipient-agent, date, time, etc. etc., to construct automatically a timing diagram, there are opportunities to build a visualization of the MAS.
Thus, the fundamental deficiencies in the tool environment MadKit were found. MadKit meet current requirements and future plans use it to develop models of human behavior in socio-economic environment.

References

  1. Тарасов В.Б. От многоагентных систем к интеллектуальным организациям [Электронный ресурс]. – Режим доступа: http://www.burnlib.com/x/tarasov-v-b-ot-mnogoagentnykh-sistem-k-intellektual-nym-organizacsiyam/

  2. Бочкарев O. Теория коллективного поведения [Электронный ресурс]. – Режим доступа: http://www.krelib.com/files/technique/Poved_Bosch.pdf

  3. Городецкий В.И., Грушинский М.С., Хабалов А.В. Многоагентные системы [Электронный ресурс]. – Режим доступа: http://www.raai.org/library/ainews/1998/2/GGKHMAS.ZIP

  4. Меркулова Т.В., Акулов Н.В. Мультиагентный подход к моделированию экономической системы в условиях нестационарной внешней среды [Электронный ресурс]. – Режим доступа: http://www.nbuv.gov.ua/Portal/Soc_Gum/Bi/2009_4/4(2)/117-119.pdf

  5. Адылова Ф.Т. Технология грид и мультиагентных систем: новая парадигма развития телемедицины [Электронный ресурс]. – Режим доступа: http://www.nbuv.gov.ua/portal/Chem_Biol/Ujtm/2008_2/2008_2_8.pdf

  6. Задорожный В.Н., Юдин Е.Б. Мультиагентный подход в имитационном моделтровании клеточных автоматов и сетевых структур – ИММОД 2007 с 72-77 [Электронный ресурс]. – Режим доступа: http://window.edu.ru/window_catalog/files/r63711/immod-mat-2.pdf

  7. Клебанов Б.И., Москалев И.М., Бегунов Н.А., Крицкий А.В.  Мультиагентная имитационная модель муниципального образования – ИММОД 2007 с 86-90 [Электронный ресурс]. – Режим доступа: http://window.edu.ru/window_catalog/files/r63711/immod-mat-2.pdf

  8. Котенко И.В. Многоагентное моделирование для исследования механизмов защиты информации в сети Интернет – ИММОД 2009 с 38-47 [Электронный ресурс]. – Режим доступа: http://www.spiiras.nw.ru/files/conferences/IMMOD-2009/IMMOD-2009-1.pdf

  9. Гулаков В.К., Буйвал А.К., Паршиков П.А. Моделирование координации субподрядчиков путем коммуникации программных агентов с компенсацией невыгодных соглашений и передачей полезности – ИММОД 2009 с 235-240 [Электронный ресурс]. – Режим доступа:http://www.spiiras.nw.ru/files/conferences/IMMOD-2009/IMMOD-2009-1.pdf

  10. Зраенко А.С., Аксенов К.А. Разработка комплекса имитационного моделирования коалиций агентов BPSIM.KIT – ИММОД 2009 с 262-264 [Электронный ресурс]. – Режим доступа: http://www.spiiras.nw.ru/files/conferences/IMMOD-2009/IMMOD-2009-1.pdf

  11. Ивашкин Ю.А. Мультиагентное имитационное моделирование больших активных систем – ИММОД 2009 с 266-271 [Электронный ресурс]. – Режим доступа: http://www.spiiras.nw.ru/files/conferences/IMMOD-2009/IMMOD-2009-1.pdf

  12. Грач Е.Г. Модель предприятия как интеллектуальная искусственная система для анализа и управления на основе знаний [Электронный ресурс]. – Режим доступа: http://masters.donntu.ru/2006/fvti/grach/diss/index.htm

  13. Лямин Р.В. Многоагентная система обучения студентов на кафедральном уровне [Электронный ресурс]. – Режим доступа: http://masters.donntu.ru/2008/fvti/lyamin/diss/index.htm

  14. Зайцев И.М Модели коллективного поведения интеллектуальных агентов в многоагентных системах моделирования и управления предприятием [Электронный ресурс]. – Режим доступа: http://masters.donntu.ru/2009/fvti/zaytsev/diss/index.htm

  15. Терзі І. Аналіз і розробка механізмів координації й співробітництва для агентно-орієнтованих систем [Электронный ресурс]. – Режим доступа: http://masters.donntu.ru/2005/fvti/terzi/diss/index.htm

  16. Зудикова Ю.В. Оценка эффективности многоагентного моделирования систем с распределенным интеллектом [Электронный ресурс]. – Режим доступа: http://masters.donntu.ru/2010/fknt/zudikova/links/index.htm

  17. Евдокимов А.А. Разработка и исследование программных агентов для распределённых вычислений в Грид среде [Электронный ресурс]. – Режим доступа: http://masters.donntu.ru/2009/fvti/evdokimov/diss/index.htm

  18. Гаврик С.С. Разработка системы управления командой роботов-футболистов [Электронный ресурс]. – Режим доступа: http://masters.donntu.ru/2006/kita/gav/diss/index.htm

  19. Блог – Агенты  и мультиагентные системы [Электронный ресурс]. – Режим доступа:http://agentlab.ru/confluence/pages/viewpage.action?pageId=6619475

  20. Т.В. Солодуха Мультиагентные системы в экономике [Электронный ресурс]. – Режим доступа: http://2010.it-edu.ru/docs/Sekzii_4-5/20_Soloduha_1253475363569911.doc

  21. Jacques Ferber, Olivier Gutknecht, Fabien Michel From Agents to Organizations: an Organizational View of Multi-Agent Systems [Электронный ресурс]. – Режим доступа: http://www.lirmm.fr/~fmichel/publi/pdfs/ferber04ocmas.pdf

  22. Jacques Ferber, Olivier Gutknecht, Fabien Michel The MADKIT Agent Platform Architecture [Электронный ресурс]. – Режим доступа: http://www.madkit.net/documents/others/MadkitTechnicalReport.pdf

  23. Muhammad Sohail MADKIT (Multi-Agent Development Kit) : A generic multi-agent platform [Электронный ресурс]. – Режим доступа: http://perso.limsi.fr/jps/enseignement/examsma/2005/1.plateformes_2/SOHAIL/SOHAIL.htm

  24. Лукина Ю.Ю., Федяев О.И. Технология создания мультиагентных систем в инструментальной среде MadKit

Note

When writing this master of the abstract dissertation is not completed yet. The final completion are December 1, 2011. Full text of the work and materials on theme can be obtained from the author or his supervisor after that date.