Online Scheduling For A Batch Processing Machine With Compatible Job Families

Recently,the online scheduling problems have received great attention.For improving the productivity in semiconductor manufacturing,metal casting et al.,online scheduling the batch processing machine with compatible job families is imperative.We mainly study the online scheduling problem that the batch processing machine serving compatible job families in this dissertation,and the following work has been done:(1)We first model the system that the batch processing machine with compatible job families as a continuous-time Markov decision process according to the work mechanism of the system,where only the information of jobs in the system is used when making decisions.Jobs from different job families can be served simultaneously in the batch processing machine,which adds the complexity while batching.Thus,we introduce some prior knowledge to simplify the problem.Based on the model we consider,we derive the optimal scheduling policy under average criterion by the policy iteration.In addition,solving the continuous-time Markov decision process by policy iteration is inefficiency if the system is large-scale.Hence,a model-free algorithm,a Q learning algorithm combined with simulated annealing is adopted to solve the problem.Simulation results finally show that the proposing scheduling method works well in the system with different parameter settings,which proves the model is rational and the scheduling method is effective.(2)There is a look-ahead range in some systems that the batch processing machine with compatible job families,where the information of jobs coming soon can be gotten exactly.In this case,scheduling on the basis of jobs in the system merely is "myopic".Accordingly,we propose the scheduling method considering future arrival information for online scheduling on compatible job families with look-ahead range,the scheduling method is called look-ahead scheduling method.The system is modeled as a semi-Markov decision process firstly.Similarly,policy iteration and the Q learning algorithm combined with simulated annealing are adopted to solve the problem.Simulation results finally show that the look-ahead scheduling method performs better than the scheduling method only using the information of jobs in the system.