Coordinate Control For Multiple CSPS System Based On State Aggregation And Demand-driven

For the coordinate control problem in a multiple CSPS system, the system statespace is composed of multiple stations’ state space whose size depends on thecapacity of the buffer. Consequently, the system state space will grow exponentiallyor geometrically as the number of stations and the capacity of buffer increase. As aresult, the learning process will suffer from the curse of dimensionality, which mayhave a negative influence on the convergence speed and optimized value. Therefore,by combining a local information interaction mechanism among stations, weintroduce a state aggregation method to reduce the size and complexity of eachstation’s leaning space. Firstly, each station is regarded as an independent learningagent that incorporates only the buffer state of its nearest downstream station intoits own learning process. Secondly, the original state space is divided into severaldisjoint sets and each set is represented by an abstract state, and a multiple-agentstate aggregation feedback Q-Learning (SAFQL) algorithm is proposed afterwards.Through our proposed approach, the agent learns the look-ahead policy over theabstract state space. The simulation results demonstrate that, in comparison togeneral feedback Q-learning algorithm, SAFQL algorithm can not only fasten theconvergence speed, but also improve the processing rate.In order to realize lean production, just in time or meet the practical demand, thisthesis incorporates a production idea which on the basis of demand driven into themultiple CSPS system and proposes a multiple CSPS production model based ondemand-driven. In the thesis, the system physical model and mathematic model arefirst built, and the joint state which is composed of the vacant capacity of bufferand the vacant capacity of bank is considered in the mathematic model. Then weaddress a multi-agent leaning algorithm based on demand-driven which is calledWolf-PHC algorithm. The final simulation results show that, through the proposedmethod, the system can obtain good optimization results and makes the productionmode of demand drives production and production promotes demand come true.