多目标差分进化算法及其在冷连轧轧制规程优化中的应用

发布时间：2018-05-01 20:00

本文选题：轧制规程 + 多目标优化　；参考：《沈阳工业大学》2017年硕士论文

【摘要】：轧制规程的制定是钢材生产过程中最重要的一环。在制定轧制规程时,需对轧制能耗、轧制功率平衡、板形板厚等诸多因素加以考虑。单目标优化虽然能很好的解决由传统轧制规程制定带来的诸多弊端,但是由于其考虑目标单一,具有较强的局限性,无法兼顾多个目标。多目标进化算法可以从多个角度出发,使得轧制规程多目标优化有着其它方法不可比拟的优势。本文对差分进化算法以及其在轧制规程优化中的应用展开研究,主要内容如下:(1)简述了轧制规程优化及其优化模型的建立方法;阐述了差分进化算法的基本原理和其常用的变异、交叉策略。(2)针对自适应跨代差分进化算法在初始阶段两个种群相同,无法有效利用父代与子代间进化信息的问题,提出了基于对位学习的自适应跨代差分进化算法。通过在算法初始阶段加入对位学习机制,使得初始化种群可以迅速找到更逼近于Pareto前沿的子代种群,使算法自始至终都可以高效的利用代与代之间的进化信息,提高了算法的收敛速度。针对冷连轧中薄窄料的特点,建立了轧制规程多目标优化模型,并用基于对位学习的自适应跨代差分进化算法对其优化。相比于实际轧制规程,优化后的轧制规程方案使得三个目标函数都得到了下降。(3)针对差分进化算法中变异操作具有很强的随机性,提出了基于进化方向的差分进化算法。该算法通过计算差分矢量在目标函数中与横轴的夹角,来选取符合收敛方向的差分矢量参与变异操作。该方法改变了差分进化算法中变异操作由随机方法生成,并通过两个多目标测试函数对算法的效率进行了验证。(4)针对轧机轧制过程中工作辊表面磨损带来的模型中参数具有不确定性的因素,直接影响轧制规程优化后解的鲁棒性,建立了考虑工作辊辊径不确定性的轧制规程鲁棒多目标优化模型。应用基于进化方向的差分进化算法对轧制规程鲁棒模型进行多目标优化,其解的鲁棒性要好于相同目标下不考虑工作辊辊径不确定性因素的轧制规程多目标优化解的鲁棒性。
[Abstract]:The formulation of rolling schedule is the most important part in steel production. Many factors, such as rolling energy consumption, rolling power balance, plate thickness and so on, should be considered in making rolling schedule. Although the single objective optimization can solve many disadvantages brought by the traditional rolling schedule, but because of its single objective, it has strong limitations, so it can not give consideration to multiple goals. The multi-objective evolutionary algorithm can be used from many angles, which makes the multi-objective optimization of rolling schedule have an incomparable advantage over other methods. In this paper, the differential evolution algorithm and its application in rolling schedule optimization are studied. In this paper, the basic principle of differential evolution algorithm and its common variation are expounded. The crossover strategy. 2) aiming at the problem that the two populations are the same in the initial stage of adaptive cross-generation differential evolution algorithm, it is unable to utilize the evolution information between father and offspring effectively. An adaptive intergenerational differential evolution algorithm based on pair learning is proposed. By adding a pair learning mechanism in the initial phase of the algorithm, the initialized population can quickly find the offspring population that is closer to the Pareto frontier, and the algorithm can efficiently utilize the evolutionary information between generations from beginning to end. The convergence rate of the algorithm is improved. According to the characteristics of thin and narrow materials in cold continuous rolling, a multi-objective optimization model of rolling schedule is established and optimized by adaptive cross-generation differential evolution algorithm based on adjoint learning. Compared with the actual rolling schedule, the optimized rolling schedule scheme makes the three objective functions drop. 3) in view of the strong randomness of mutation operation in the differential evolution algorithm, a differential evolutionary algorithm based on evolutionary direction is proposed. By calculating the angle between the difference vector and the transverse axis in the objective function, the differential vector in the direction of convergence is selected to participate in the mutation operation. This method changes the mutation operation in differential evolution algorithm, which is generated by random method. The efficiency of the algorithm is verified by two multi-objective test functions. (4) aiming at the uncertainty of the parameters in the model caused by the wear on the surface of the work roll in the rolling process of rolling mill, the robustness of the optimization solution of the rolling schedule is directly affected. A robust multi-objective optimization model for rolling schedule considering the uncertainty of work roll diameter is established. The multi-objective optimization of rolling schedule robust model based on evolutionary direction differential evolution algorithm is better than that of rolling schedule multi-objective optimization solution without considering the uncertain factors of work roll diameter under the same objective.
【学位授予单位】：沈阳工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TG335.13;TP18

【相似文献】