Research On Scheduling Algorithm For Semiconductor Production Line Based On EDA And CS

Today's semiconductor manufacturing industry is often faced with complex,multiple entry,large-scale and highly uncertain production conditions.The heuristic method can solve its scheduling problem quickly,but it has a strong dependence on the scheduling environment and goals,and it has poor generality.In addition,it is difficult to guarantee the quality of solutions.Intelligent computing methods can theoretically yield high-quality solutions,but the computational complexity of these algorithms generally increase significantly with the increase in the problem size and constraints.Therefore,the scheduling method that takes into account effectiveness,efficiency,and real-time has theoretical and economic value.This paper aims at the semiconductor manufacturing process.In order to obtain high-quality scheduling solutions with lower computational complexity.Based on the difficulties in the scheduling process and intelligent scheduling methods,we draw lessons from machine learning theory and mathematical thinking,and study an high-efficiency intelligent algorithm based on estimation of distribution algorithm and cuckoo search algorithm for semiconductor production line.The specific research content is as follows:1.Consider the actual situation that the job in the semiconductor production line should wait as little as possible in the partial machine buffer,we give an unrelated parallel machine scheduling problems with constraint waiting time.For this type of scheduling problem,an estimation of distribution algorithm based on Coupla theory is studied.This algorithm uses the Coupla theory to construct a joint distribution function for each machine based on the ratio of the number of similar order items to the total number of jobs,and then establishes a probabilistic model of dominant populations.The progeny encoded vector sets obtained by sampling the functional model of the joint probability distribution retain the relative position information of the parent advantage code.Theoretically analyzing the time complexity of the proposed algorithm,it increases logarithmically with the number of jobs,and is therefore suitable for large-scale practical production scheduling problems.2.In order to reduce the makespan of semiconductor final testing phase,a scheduling method that combines reinforcement learning,surrogate model,and cuckoo algorithm is studied.This method uses a cuckoo search algorithm as a scheduling method framework.In order to balance diversification and intensification of population of cuckoo search algorithm,an offline training parameters model based on reinforcement learning technology is introduced,and the parameters are adaptively adjusted online;in order to speed up the algorithm search process,reduce a large amount of computation that is brought about by calculating the fitness function value for many times in the process of off-line training model and on-line optimization of the cuckoo algorithm,the relative ranking of solutions are estimated by surrogate model technology.In order to achieve a better solution at the same time as shortening the overall search process of the evolutionary algorithm.Simulation experiments show that the Coupla theory-based estimation of distribution algorithm and fusion reinforcement learning,agent model and cuckoo algorithm scheduling algorithm studied in this paper can effectively balance the quality of the solution and the solution time,and has a certain industrial application potential.