Theoretical Aspects of Knowledge base Quantization in Multi-Agent Systems
Abstract
Introduction: Intensive use of wireless and mobile communication networks in distributed information-telecommunication systems and automated systems of control over technological processes in industrial corporations leads to the situation when the architecture of such systems forms new conceptual entities called agents which should be controlled in real time. When assessing the effectiveness of multi-agent systems, you should take into account the communication complexity of the interaction between the agents, which directly depends on the number of the adapted parameters. For intelligent agents, the main issues in this case are, on the one hand, the requirement of rapid adaptation and self-adaptive parameters of their knowledge bases and, on the other hand, the need to ensure the integrity and security of both the data and the agents' knowledge. Purpose: The goal of this research is developing theoretical and algorithmic ways for soft quantization of additive production knowledge bases, which would reduce the communication complexity of the interaction between intelligent agents. Results: We have proved a number of statements which allow you to perform isomorphic transformations of classical additive fuzzy models functioning in the field of real numbers into their analogs able to function in finite Galois fields. We have demonstrated the homeomorphism (topologic isomorphism) of the discussed fuzzy and neuro-fuzzy models. On the base of the proven statements, an algorithm of soft quantization of fuzzy implications has been developed. The essence of this algorithm is flexible control over the topology of neuro-fuzzy hierarchical models while maintaining the given accuracy of the approximation of the agent-controlled parameters. The novelty of our approach is the substantiated principal possibility of adequate quantization of additive production knowledge bases. A way has been found to automatically control the amount of knowledge base optimization parameters which include the number of terms in linguistic variables, the number of adaptation parameters for each linguistic terms, and the number of hierarchical levels in an additive fuzzy model. As constraints, we have accepted the assumptions about continuous coverage of the solution space (continuum) by terms of linguistic variables of fuzzy implications. It has been shown that to ensure a quantization continuum, it is sufficient to adjust the Galois field characteristic. Practical relevance: The developed method for soft quantization of fuzzy implications allows you to significantly (in several orders) reduce the volume of the knowledge base necessary for the agents to make appropriate decisions, thereby reducing the communication complexity of the interaction between the agents. Also, it provides the possibility to adjust the level at which the data and the knowledge of the agents are protected.Published
2017-06-21
How to Cite
Fomicheva, S. (2017). Theoretical Aspects of Knowledge base Quantization in Multi-Agent Systems. Information and Control Systems, (3), 2-10. https://doi.org/10.15217/issn1684-8853.2017.3.2
Issue
Section
Theoretical and applied mathematics