Research On Scheduling Methods Of Complex Industrial Process Based On Machine Learnnig Nad Swarm Intelligence Algorithms

Scheduling of complex industrial process is a focused area in the field of manufacturing science,operational research,and industrial engineering,etc.It is also a key technique to ensue an efficient operation of production processes and improve energy saving.With the help of effective production scheduling,enterprises can obtain higher return on investment without increasing too much input.With the rapid development of technology and industrial production,the current development of manufacturing industries is changing to customized production mode with large-scale,mulita-batches,and small batch size.The probability of dynamic factors in manufacturing process increase at the same time,and it becomes more and more difficult to establish an accurate mathematical model to analyze,control and schedule manufacturing processes based on mechanism.Operational methods are difficult to find an optimal solution in an acceptable time and require ideal hypothesis.Herustic rules rely heavily on environment,thereby limiting the quality of schedules.The computational intelligence are parameter-sensitivity and time-consuming when solving a high dimension problem.Therefore,research on scheduling problems of complex industrial processes has significant application meaning,yet of extremely theoretical challenging.This paper focuses on the following four realistic issues from real-world scheduling problems: performing objective function evaluation,tuning parameters in a dynamic environment,balancing computational cost and degree of confidence,and solving a high-dimensional optimization problem.To the above mentioned difficulties,this work combines the theory of artificial intelligence and the basic idea of operational research,and studies intelligent scheduling methods for complex industrial processes based on machine learning and swarm intelligence algorithms.Details are as follows:1.Considering the batch scheduling problem of a diffusion and oxidation zones in semiconductor production line,this paper studies a two-level optimization framework based on problem decomposition.It divides the original problem into a batch formation sub-problem and a batch scheduling one.The former is solved by using a time windows strategy with decision theory.The latter is optimized by a surrogate-assisted symbiotic organisms search algorithm.In the later,a reinforcement learning-based parameter tuning method is constructed to balance the global and local search of symbiotic organisms search algorithm.The surrogate model,which can predict the sequencing result instead of time-consuming true fitness evaluation,is used to reduce the computational burden of symbiotic organisms search algorithm and the training process of reinforcement learning.2.Consider the scheduling problem of semiconductor final testing stage,this paper studies a learning-based grey wolf optimizer.In order to handle with the parameter-sensitivity problem of grey wolf optimizer,a reinforcement leaning method with a developed delay updating strategy is used to build a parameter tuning scheme.The scheme constructs a reward function that aims to balance the global and local search of grey wolf optimizer.It is off-line trained and update online,thereby reducing the computational burden greatly.The proposed algorithm can significantly improve the quality of a solution and while saving computational time.3.Consider stochastic parallel machine scheduling problem in a real-world manufacturing context,in which its processing time can be described by a gamma or log-normal distribution.In order to obtain a high reliable schedule,an optimal computing budget allocation strategy is proposed.By making full use of both prior distribution and simulation results,the proposed strategy can intelligently determine the computing budget,and maximize the probability of correction selection of the best schedule.The theoretic interpretation of the proposed optimal computing budget allocation strategy proves,for the first time,that it is a global optimum of the consider problem.4.Consider an extended flexible job shop scheduling problem from a postprinting environment,this paper uses a directed acyclic graph to describe the precedence among operations,and formulates its a mixed integer programming model.A learning-based cuckoo search algorithm is proposed to solve the considered problem.In it,an improved one-hot encoding method is proposed to map the population of cuckoo search into a European space.A sparse autoencoder is introduced to compress a high-dimensional solution into an informative low-dimensional one.In order to reveal the linkages among decision variables and enhance the explore ability of the proposed method,a factorization machine is used,for the first time,to capture the relevant and complementary features of population.Hence,a parallel framework involving low-level and high level populations is constructed.The developed method fits to handling high-dimensional optimization problemsNumerical simulations with benchmark and randomly generated ones show that using the proposed method spend less time while obtaining a schedule with good quality,strong-robustness,and high confidence level.Hence,they can be readily applied to industrial scheduling problems.