Research On Multi-objective Evolutionary Algorithm And Its Application To Rolling Schedule Optimization

Many objectives need to be considered at the same time when doing rolling schedules optimization. These objectives are rolling energy consumption, balance of power distribution and so on. Therefore rolling schedule optimization requires optimizing multiple conflicting objectives simultaneously in the highly complex space. The conventional multi-objective optimization methods can not solve such problems effectively. Multi-objective optimization evolutionary algorithms have many unmatched advantages compared with the conventional multi-objective optimization methods in solving multi-objective optimization problems. However the shortcomings of the multi-objective optimization evolutionary algorithms themselves and the application difficulties are big blocks for their application in rolling schedule optimization problems. The research works of this article aim at multi-objective evolution algorithms and their application in rolling schedule optimization. The main contents and results are as follows:1) A brief review of multi-objective evolutionary algorithms and rolling schedule optimization problems is made on basis of searching many literatures, which contains their basic conceptions, development and research status quo.2) In order to enhance search ability and expedite convergence rate of the multi-objective optimization evolutionary algorithm, which makes it suitable for solving rolling schedule optimization problems, an opposition learning based multi-objective genetic algorithm is proposed. The algorithm combines opposition learning with non-dominated sorting genetic algorithm, and adopts opposition learning in the initialization and evolutionary process of the population. As a result the performance of NGSAâ…¡is improved. Then the application of the proposed algorithm in rolling schedule optimization of hot finishing rolling is researched. The multi-objective optimization model is established according to characteristic and rolling process mathematic models of hot finishing rolling. Opposition learning based multi-objective genetic algorithm is applied to do rolling schedule optimization of hot finishing rolling and better rolling schedules are obtained compared with original schedule.3) In order to enhance local search ability and solving efficiency of the multi-objective optimization evolutionary algorithm, an adaptive local search algorithm which is suitable for single objective differential evolutionary algorithm is extended to multi-objective evolutionary algorithm. The key problems of extending are solved by adopting clonal selection operator for the individuals updating. Meanwhile a method to change the clonal scale of different individuals adaptively is proposed. Then a clonal selection based adaptive local search algorithm is formed. A clonal selection adaptive local search based multi-objective Memetic algorithm is proposed by combining clonal selection based adaptive local search algorithm with multi-objective genetic algorithm. The idea of single objective evolutionary algorithm convergence proof is extended to multi-objective evolutionary algorithm. After that convergence of the proposed algorithm is proved. Then the application of the proposed algorithm in rolling schedule optimization of hot finishing rolling is researched. Simulation shows that better rolling schedules are obtained with less running time compared to original rolling schedule and opposition learning based multi-objective genetic algorithm, which laid the foundation for online application of multi-objective evolutionary algorithm based rolling schedule optimization.4) In order to reducing the solving time of rolling schedule multi-objective optimization problem, and increasing probability of its online application, high quality Pareto solution set needed to be obtained using small population. So an immune algorithm based normalized normal constraint method is proposed combining immune algorithm with normalized normal constraint method. The convergence of immune algorithm based normalized normal constraint method is analyzed. Then the multi-objective model of rolling schedule is established according to actual rolling schedule optimization process. Good optimization results are obtained by simulation test of immune algorithm based normalized normal constraint method.5) Aiming at further reducing the solving time of rolling schedule multi-objective optimization problem, an improved adaptive weight approach is proposed, and it is applied to rolling schedule multi-objective optimization of tandem cold rolling based on GA, PSO and clonal selection algorithm. Additional objective function which reflects slip rate of the strip is added to the original multi-objective optimization model of tandem cold rolling. Applying the improved adaptive weight approach GA, PSO and clonal selection algorithm to optimize rolling schedules of different specifications strips, satisfied results are achieved. The improved adaptive weight approach evolutionary algorithm based rolling schedule optimization method can optimize more objectives, use less runtime and are more likely to be used online.