Optimization Control Approach Of CSPS Based On Event And Stochastic Demand

In many real-world production lines, there is a production system which is mainlycomposed of a production station; such a system is called conveyor-serviced productionstation (CSPS). It is a very important abstract model in real-world practical production,which appeared in the industrial production automation and was used to the implementationof production automation. The research of optimal control problem of CSPS system byfinding out an optimal control policy to reduce the production wastage rate, obtain theoptimal long-run expected cost of the system and improve production output is of greatfeasible significance.The look-ahead control of CSPS system is based on the vacant of the buffer. Thesystem makes a decision according to that vacant. The unloading or processing a part canmake the same vacant of the buffer, so the thesis gives the natural event. Otherwise, thevacant of the buffer can be clustered by some character such as measuring off a fix value,so the thesis gives the clustering event. The thesis discusses two different events, which arethe natural event-based optimization control model (NBEO) and the clustering event-basedoptimization control model (SCEBO). In GBEO the same state is divided into differentevent, which can increase the event space and also the precision of learning. In SCEBO theCSPS system maps the vacancies of the buffer into some clustering events, which arefully-full event, semi-full event, semi-vacant event, and fully-vacant event. It can reducethe event space and improve the learning efficiency of the system. Finally, the thesis givesthe unified event-based Q-learning and online policy iteration in CSPS according to thesetwo different events.The research of CSPS system sets the bank which is set to be infinity. This is differentfrom the real word. So this thesis introduces the optimization design and control problem ofCSPS based on stochastic demand, in which the bank of the system is set to be finite. Thethesis discusses the optimization design of CSPS based on stochastic demand by designingthe capability of buffer and bank. Finally, the thesis gives the Q-learning algorithm basedon simulated annealing.