Current expert systems are applied to limited domains with small quantity of knowledge for problemsolving. In the future, we will face the problems that consist of many subproblems and require much knowledge to solve. Distributed problem solving systems,consist of many cooperative problem solvers (we call 'agents' in this paper),are effective tools for such problems. In this paper, we define problem-solving as a search from initial state to goal state, and mention about a distributed search mechanism. Since the control of the system is distributed,there are some problems for coherent search. (1) It is tend to make a useless search caused by message delay. (2) It is hard for each agent to check its solution with the best one. (3) It is hard to cut search-path globally. To solve these problems,we suggest a distributed search mechanism using 'priority token passing protocol'. By this protoco,(1) it is able to request an agent to search according to the importance,using' request search' message. (2) It is easy to check a solution with the best one using 'confirm end' and 'demand solution' messages. (3) It is able to cut search-path globally using 'cut path' message.
抄録全体を表示