| In modern mass-production industry, processing jobs in batch could effectively improve the production rate, and thereby meet the requirements of modernization. Therefore, research on the optimal control of batch processing system is an interesting topic in production optimization area, which has both theoretical and practical significances.In this thesis, we propose a control model for conveyor-serviced batch processing system by introducing predictive control mechanism. In the model, machines that have the ability of processing jobs in batch are able to acquire specific information of the jobs by looking some distance ahead with the sensors deployed. Under the cases of both size-different and service-time-different jobs, online optimal control of the system with a single machine is mainly concerned. Optimization model of the system is built by using look-ahead range as control variable. The objective is to maximize the job processing rate of the system in infinite horizon. Two decision processes are included in the model, which are batching process and action-selection process. For the batching process, the rule of minimizing process-capacity wasting ratio of the machine during production period is proposed. And for the action-selection process, model-free Q-learning algorithm is employed to derive the optimal look-ahead policy.We further study the online optimal control of demand-driven batch processing system. Firstly, by using infinite-horizon profit losing rate as optimization objective, we build a real-time dynamic batching model for the system. Then, a Q-learning based online optimization algorithm is presented to derive the optimal control policy.Finally, we show the effectiveness of our proposed model and online control algorithm by conducting several simulation experiments, and analyze the effects of some batching rules and physical parameters on system performance. |
PDF Full Text Request