29 Август 2013
The complexity of dynamic systems based on knowledge
Systems theory is a rapidly developing field of science and practice shows that the main problem in the theory of complex systems is the problem of decision making in the presence of many goals.
A dynamic system should be considered a system that has such properties as connectivity, complexity, stability, and goals of behavior, which are poorly formalized.
Modern information subject areas are the Internet, prediction of accidents and emergencies, distributed learning, etc.
Their features are as follows: the presence of a huge number of autonomous entities with their specific subgoals (autonomy), entities are subject to the influence of the external environment (openness), entities interact with each other (distribution), the knowledge bases of entities are unique (locality), entities form hierarchical coalitions (hierarchy of levels) .
To build models of such subject areas, for example, self-organizing open multi-agent systems are used.
To study the global properties of dynamic processes in subject areas, mathematical structures from differential topology are used - manifolds, which are interpreted as a "state space", and the dynamic process is modeled using differentiable mappings (diffeomorphisms or smooth flows).
A mathematical model of a dynamic subject area is constructed in the form of a hierarchical functional system (FS), in which the knowledge base is associated with a chain of bundles of knowledge bases, i.e. is a section of a chain of bundles of knowledge bases.
FS is characterized by the following properties:
connectivity - a chain of bundles of the knowledge base;
complexity - the hierarchy of levels of local knowledge bases;
stability (dynamic behavior of the system) – the structure of the section of the chain of bundles does not change when the local knowledge bases of the bundled manifold are perturbed in the vertical direction. In other words, only the heuristics of the local knowledge bases of the bundle chain are changed, and the bundle base, which is interpreted as the external environment, remains unchanged.
A computer model of a hierarchical FS is implemented, which is a formalized reflection of the subject area in the form of a hierarchical structure of a set of control components (agents) that interact with each other to achieve the main goal.
The technical implementation of a computer model of a hierarchical functional system is made in the KARKAS tool environment, which is designed to build computer models of multiply connected systems based on bundles of knowledge bases for various purposes: economic, environmental, informational, etc.
The software implementation of the "KARKAS" system modules is based on the use of client-server technology based on socket programming.
In this case, the system implements client-server interaction: "thick client" - "thin server", i.e. the server part implements only access to system resources (identifying a student, establishing a connection with him and receiving test scores from him), and the main part of the application (inference engines, explanation, training, knowledge base) is located on the client. In other words, the system works both as a server and as a client, i.e. is a software agent.
Depending on the problem statement, the agent has three modes of operation:
- development of the knowledge base: formation of heuristics of the subject area, debugging the knowledge base (expert mode);
- testing the user's knowledge in the local network (user mode), for example, is carried out as follows: one of the network computers is declared the teacher's workplace and the agent works as a server, and on the other computers the agent works as a client;
- distributed learning (collective communication mode), i.e. agents (clones of the KARKAS system) are located in the network (the IP address of the computer is determined automatically) and can function independently. In this case, the user can individually download the knowledge base and perform a consultation at a convenient time for him and communicate with other clones of the system on the network.
