Nondeterministic Observation Reduction In Multi-agent Domain

Observe the information reduction is a frontier field of intelligent planning. Thecurrent research on observation information reduction abroad are mainly focused onplanning for the single agent solution, and some problems need to be studied toobserve the information reduction, especially in the area of multi agent planning,solving the agent collaborative planning solution needs to consider the relationshipbetween each agent, so that they are in competition and conflict situations next, theachievement of their own or collective goal. How collaborative planning wereobserved information reduction in agent, improve the efficiency of cooperation ofmulti agent and reduce the execution cost. Is a problem worthy of study.Aiming at the uncertain multi agent planning system, discusses the problems inthe field of multi observation information reduction in agent programming. Thisarticle absorbs the basic thoughts of uncertain information reduction plan, it can beextended to the field of multi agent. This paper first presents the minimum observedmulti agent planning problem. This paper presents a distinguishable matrix, torepresent the observed variables and the need to distinguish between the state of thecorresponding relationship. Distinguish matrix based, this paper establishes a set ofalgorithms, through the branch and bound method to search the minimum observeddistinguishable matrix Jane path different to solve multi agent planning. As proved bythe algorithm application and theory, which indicates that the algorithm can find thesolution, and find the solution is consistent with the optimal requirements. The timecomplexity of the algorithm is not higher than O (2mkm) of not less than (km2), wherek denotes the number of the state should be distinguished, M represents the number ofobserved variables.Based on the minimum set problem on observation of extended forward and adetailed definition of the optimal multi agent planning field observation set problem.And put forward the weighted distinguishable matrix to reflect the observed variablesand the need to distinguish between the state of the relationship between. And to solvethis problem, a hybrid algorithm for solving the problem and design a geneticalgorithm with the greedy method. Through experiment and theory analysis, thealgorithm can effectively solve the problem.