Модель целеполагания в многоагентной системе с ограниченным ресурсом времени

Abstract

The article considers the goal-setting model concept in modeling the reasoning of cognitive agents in dynamic hard real-time systems. An important issue concerned with processing tasks by means of hard real-time systems is the ability to estimate the amount of time available for processing till the moment of time it becomes too late to produce the result. To avoid such a situation, it is necessary to be able to correlate the steps and results of ongoing reasoning with the events occurring in the external environment. Reasoning of this type is called reasoning in time. To formalize such reasoning, formalisms of active logic are suggested. When regarded as a deduction model, active logic is characterized by a language and a set of deductive rules and observations. By using an observation function, one can model a dynamic environment, information about which arrives to the agent as changes occur in that environment. Reasoning in time is characterized by accomplishment of deduction cycles (steps) serving as a time standard. The agent's knowledge is associated with the index of the step on which it was gained for the first time. Observations can be carried out at any step of the deductive process, and a formula expressing some statement and associated with the relevant step serves as its result. To construct a goal-setting model for solving problems in hard real-time environment, a logic system (a metalogic of knowledge) is introduced to formalize in the general case the metareasoning of several intellectual agents about the tasks they face, about their own knowledge, and about the possibilities of solving these problems in interaction with the other agents, taking into account the limited time resource available to each of the agents. The goal-setting process gives an answer to the question of which subgoals and in what sequence at a given moment in time should be achieved for the main goal faced by the multi-agent system be fulfilled within acceptable time. It should be noted that the response may vary depending on the current situation and as additional information becomes available, because the assigned targets may be canceled and new ones assigned. The knowledge metalogics language is shown as a combination of three languages: the language of problem domain, the language of agents, and metalanguage, with the last two being multigrade ones. The language of agents contains special predicate symbols for reasoning about goals, about the tasks associated with them, about the structure of these tasks, and about possible methods and admissible timeframes for solving them. Metalanguage allows one to express general statements about the arguments of agents. Formal theories constituting the meta-knowledge system of knowledge are presented.