Research On Semiconductor Parallel Machines Scheduling Based On Surrogate Model Simplification

Nowadays facing semiconductor manufacturing line complex processing of environmental uncertainty,carry out data driven production process modeling,the modeling and forecasting of data mining to extract effective information,guide the production process and provide the best solution.Considering many practical scheduling problems,such as parallel machine scheduling,batch scheduling,according to the idea of decomposition of complex scheduling problems can be simplified into several hierarchical inheritance sub problems,but because there is a strong coupling relationship between the upper and lower decision,there is a lack of long time calculation and low efficiency of the intelligent evolutionary algorithm for scheduling.Therefore,it is difficult to apply to the actual algorithm simulation and optimization of operation process.In order to solve the contradiction between computational time efficiency and solution accuracy,fast approximation instead of time-consuming evaluation process technology surrogate model,take the rough estimate and the accurate evaluation of the combination of the way,under the premise of ensuring the quality of solution,accelerate the evolution of the whole algorithm process,realize multi-stage synchronous optimization task.In this paper,we focus on the scheduling algorithm of semiconductor parallel machine:(1)This paper describes and analyzes the basic workflow,the basic properties of the workflow and the complex process constraints of the parallel scheduling and parallel batch scheduling.In depth study on how to allocate the equipment,work scheduling,and the specific operation relationship between batches in dynamic scheduling decision making,and propose a hierarchical algorithm for solving different scheduling scenarios.(2)According to the production line of history storage and real-time information collection data on the key performance characteristics closely related to scheduling problems,the establishment of a variety of agent model has the ability of forecasting,and based on the model evaluation standard models,fast and efficient evaluation of optimal target value.In order to avoid the deviation of the search results caused by the model error,the accuracy of the optimal scheduling solution is guaranteed by the combination of the exact evaluation and the updating of the model.(3)The organic combination of technology and intelligent agent model evolutionary algorithm,is studied based on evolutionary algorithm as the framework of the upper equipment allocation,using approximate predictor,simplified algorithm to find the optimal lower order decision-making time-consuming link,so as to stimulate the local exploration of new potential to reduce computational complexity with nested structure algorithm,and improve the overall algorithm convergence.Combined by the proposed model evolutionary algorithm with the traditional scheduling method is compared,through a lot of simulation results show that this method has the ability to obtain better quality of the solution,and speeds up the search algorithm,which improves the efficiency of algorithm,and realize the collaborative optimization of multi stage and multi hierarchy decision,provides a novel way to solve complex scheduling problems of semiconductor.