An agent-oriented approach to the organization and development of distributed intelligent systems

Abstract

   Objectives. The work set out to justify the expediency and provide a conceptualization of an agent-oriented approach to building distributed intelligent systems that operate under conditions of incomplete initial data, high stochasticity, and non-stationarity of the external environment. Formal models of agents and methods for their cooperation were developed as part of a decentralized architecture that ensures adaptability, scalability, and fault tolerance. A distributed knowledge storage and processing subsystem was designed and implemented.    Methods. In order to solve the problems, agent modeling methods were integrated with automata theory, Markov decision processes, fuzzy logic, artificial neural networks, multi-agent reinforcement learning algorithms, along with expert evaluation procedures to ensure the verification and practical adaptation of the developed theoretical provisions.    Results. The developed multi-agent intelligent system architecture based on the cooperation of autonomous software agents provides decentralized control, weak coupling of components, and scalability of the computing infrastructure. The proposed formal mathematical model of an intelligent agent integrates symbolic knowledge representation methods and neural network algorithms. A distributed knowledge base management subsystem was developed to provide dynamic reconfiguration of information flows, adaptive optimization of data storage and processing schemes, intelligent caching, semantic consistency, and support for non-monotonic distributed knowledge-based inference. A dynamic load balancing subsystem was implemented based on a coalition of system software agents.    Conclusions. The described agent-oriented paradigm is an effective methodological basis for building distributed intelligent systems aimed at solving problems in poorly formalized problem domains characterized by incomplete data, uncertainty, and non-stationary external environments. Decentralized management and distributed knowledge base organization allow overcoming the fundamental limitations of monolithic intelligent systems which use the complete and consistent knowledge base and the monotonous logical inference, as well as ensuring the scalability and high fault tolerance of the developed system. The mechanisms of dynamic optimization of a distributed knowledge base and adaptive balancing of computational load implemented in a multi-agent intelligent system can be used not only to solve intelligent problems, but also to effectively manage their own computational and information resources.