Study On Decision Making And Knowledge Updating Mechanism In Dynamic Scheduling

The existing international economic order and structure is broken and recombined by information revolution. The economy is mainly depended on information. Informatization must be carried out because of rapid changing of market environment. Dynamic scheduling is an important topic in manufacturing informatization. The value of its theory and application are more and more outstanding. This thesis focuses on knowledge-based decision mechanism of dynamic scheduling. Firstly, a system architecture based on multi-agent is constructed, and a new dispatching strategy based on reinforcement learning is presented. Then, a Contract Net Protocol based resource distribution mechanism and the process of knowledge acquiring and fusion are studied respectively. At last, the knowledge updating is discussed.1. The architecture of the dynamic scheduling system is constructed based on multi-agent. There are three levels that are composed of various agents in the system. The main agents are introduced, and the unified agent architecture is structured. The architecture of the system is presented. The multi-agent based architecture can satisfy kinds of requirements in manufacturing enterprises.2. A job dispatching strategy based on reinforcement learning is presented. The action group is presented which contains static and dynamic actions in Q learning. In the course of selecting action group, the concept of cohesion is introduced, and a selecting strategy based on equilibrium of distribution is presented. The related concepts in job shop are given out, then the job dispatching knowledge base is bulit and reward function is designed.3. The knowledge base modeling of resouece distribution is studied. To enhance the efficiency of consultation in traditional multi-agent system, an improved modeling of CNP is presented, and the bidding set that is used for the framework of knowledge acquiring is described. The knowledge acquiring of attributes based on rough set is studied, and the framework of decision table is constructed. The knowledge acquiring of decision attribute based on the set of experts'action instances is studied. Then the decision tables are obtained and analyzed. In order to solve the conflict of different decision attributes from different experts, a knowledge fusion strategy based on balloting combine with experts'authority is presented, and then the common decision table comes into being. 4. A resources distribution mechanism based on CNP is studied. The bidding-processing algorithm is described. According to different influence of condition attributes to knowledge reasoning, the weight and reliability of condition attributes are discussed. The knowledge reasoning mechanism based on Fuzzy Petri Net is studied. This mechanism applies standardized attributes knowledge and knowledge of scheduling rules in fuzzy reasoning, and then distributes resources for each job.5. The knowledge updating in dynamic scheduling decision system is studied. The necessity of knowledge updating is illustrated both in inside and outside environment of the manufacturing enterprise. An adaptive updating mechanism based on reinforcement learning is presented. When initializing decision table, the support rate of rule knowledge is integrated, and the weight of condition attribute is considered in the reward function. The processing of scheduling decision and knowledge updating will be more reasonable and effective.6. This thesis discusses the design and implementation of dynamic scheduling decision system. Its framework is integrated on the platform of Jini technology. The functions of the modules are described, and then the system sequence diagram is given out. On basis of theories presented in this thesis, the mold system is implemented primarily.