Research On Multi-objective Evolutionary Algorithm And Its Application To Manufacturing System

Many real-world problems often require simultaneously optimize multiple competitive or conflicting objectives. Evolutionary algorithms have incomparable advantages over traditional methods, e.g. the weighted-sum approach, in solving multi-objective problems. On one hand, the evolutionary algorithms can optimize simultaneously multiple criteria without the consideration of the decision maker's preference information. On the other hand, the parallel mechanism makes the evolutionary algorithms be able to obtain the Pareto optimal set or an approximation set in a single run. The study on the multi-objective evolutionary algorithm (MOEA) has not only the theoritical value but also the practical meaning. Focused on the MOEA and its application to manufacturing systems, the topics researched in this thesis and the innovations of our investigations are listed as follows:The thesis puts forward a hybrid fine-tuned multi-objective memetic algorithm (HFMOMA). The method combines a local search procedure with recombination operators (crossover and mutation) to improve the global search ability. In order to exploit the search space, a simulated annealing (SA) is used in the local search stage in the HFMOMA to optimize each of the random linear weighted functions. In the recombination stage, the algorithm uses the Pareto method to implement the crossover and mutation, and adopts a grid-based local perturbation (GBLP) to enhance the exploration capabality of the algorithm. Besides, for obtaining a better robusticity, the optimization process of the HFMOMA can be fine-tuned dynamically and adaptively according to the online feedback improvement rate. By solving the multi-objective 0/1 knapsack problem instances, the simulation experiments show that compared with several other MOEAs, the approximations found by the HFMOMA are more close to the Pareto optimal set and with the better diversity.A SA based multi-objective memetic algorithm (SAMOMA) is designed and applied to a flowshop scheduling problem (FSSP). The innovation of the SAMOMA is the hybridization of the SA based local search, evolutionary operators and a grid-density based selection mechanism so as to enhance the global search ability of the algorithm. Moreover, the Pareto domination relationship is used to evaluate the fitness in the SAMOMA and calculate the acceptance probability in the SA. Simulation results show that the convergence speed of the SAMOMA is faster than that of other SA based MOEAs, and the approximations generated by the SAMOMA are more uniform. Besides, the thesis also discusses the applications of a dynamic periodic weighted function based MOEA and a ant colony algorithm to the FSSP.The thesis proposes a multi-objective memetic algorithm (MOMA) to solve the multi-objective sequencing problem for a mixed-model assembly line in a Just-in-time production system. The characteristics of the MOMA are using a greedy based local search and a grid density-based selection strategy to improve the exploitation and exploration of the algorithm. Simulation results on the problem instances show that the search speed of the MOMA is quicked than that of the enumeration method, and compared to a multi-objective genetic algorithm (MOGA) that is effective on solving the problem instances, the results show that the proposed method has higher search efficiency, and it reveals the evident advantage over the MOGA especially for the large-scale problem.Aiming at the internal supply chain management of semiconductor manufacturing corporations, the thesis studies the multi-objective optimization of the bin allocation planning and the multi-factory capacity planning in semiconductor manufacturing. The bin allocation planning is formulated as a transshipment problem model in this thesis and the advantage of the model is that it reduces the variables of the traditional formulation approach. For the multi-objective optimization of the bin allocation decision problem, the thesis developes a lexicographic ordering based heuristic method for the problem. Compared with the linear programming on the same problem instance, the simulation results show that the heuristic can obtain the satisfied solution without the need of deciding weights. For the multi-objective multi-factory capacity allocation planning problem, we propose a two-stage model for the problem and develop a SA based MOEA to solve the model. Experiment results show that it is feasible to deal with the multi-objective capacity allocation problem in semiconductor manufacturing by the use of an evolutionary algorithm.At the end of the thesis, we summarize all our works and put forward the future research directions.The study is supported by National Science Fundation of China(No. 60174009) and Intel/High Education Fundation of Research.