基于数据解析的连退生产过程多因子操作优化

针对连退生产过程中带钢质量波动大和生产能耗过大的问题,基于数据解析方法构建带钢质量的预测模型,进而建立连退生产过程多因子操作优化模型.该模型的任务是求得一个最优工艺参数设定方案,使得模型中所包含的两个相互影响但并不冲突的目标能够实现同时最优化.针对该问题,提出一种改进的自适应多因子进化算法(AdaMFEA),将不同优化目标作为不同类别因子,通过父代解在不同因子上的性能评价指标决定子代解的搜索方向.为了改进算法的鲁棒性和搜索效率,算法使用多种交叉算子,并基于各算子的搜索性能分析提出多种交叉算子的自适应选择机制;同时提出基于回溯直线搜索和拟牛顿法的个体学习策略,对个体进行局部搜索.基于Benchmark问题的实验结果表明,AdaMFEA能够有效提升传统多因子进化算法(MFEA)的求解效率;基于实际工业问题的实验结果表明,AdaMFEA可有效求解连退生产过程多因子操作优化问题,实现多个非冲突目标在一个种群的进化过程中同时达到最优.

Multifactorial operation optimization for continuous annealing process based on data analytics

To deal with the fluctuations of strip quality and excessive energy consumption in continuous annealing production process, this paper firstly establishes a prediction model for strip quality based on data analytics and then builds a multifactorial operation optimization model. The task of the model is to achieve an optimal parameter setting so that the two optimization objectives, which interact but do not conflict with each other, can be optimized simultaneously. To solve this model, an adapive multifactorial evolutionary algorithm(AdaMFEA) is proposed. The algorithm regards different optimization objectives as different categories of factors. The evaluation and search directions of individuals are determined by the information transmission of factors from parents. To imporve the robustness and search efficiency, multiple crossover operators are adopted in the algoirthm and an adaptive selection strategy for these operators is designed according to their search results. Meanwhile, an individual learning strategy based on the backtracking linear search and the quasi-Newton method is also proposed. The experimental results based on the Benchmark problem show that the AdaMFEA can effectively improve the efficiency of the traditional multifactorial evolutionary algorithms(MFEA). The experimental results based on practical industrial problems show that the AdaMFEA can effectively solve the multifactorial operation optimization problem in the continuous annealing production process, and achieve the simultaneous optimization of multiple non-conflicting objectives in the evolution of a population.