基于预测与分解策略的大规模炼油过程生产调度算法

Production scheduling algorithm for large-scale refining process based on prediction and decomposition strategy

中文关键词  分解算法  深度学习  大规模优化  炼油生产  调度

英文关键词  decomposition algorithm  deep learning  large-scale optimization  refinery  scheduling

中文摘要

      炼油生产调度为混合整数规划问题, 随着规模的增大, 其求解时间随问题规模呈指数增加, 使得大规模长周期炼油生产调度问题难以在合理的时间内求解. 针对该问题, 本文提出了一种基于生产任务预测与分解策略的炼油生产调度算法, 该算法能在短时间内获得大规模调度问题的满意解. 所提算法将原问题沿时间轴分解为若干个调度时长相同的单时间段子问题, 并设计了基于深度学习的单时间段生产任务(组分油产量)预测模型, 用于协调子问题的求解. 其中, 生产任务预测模型通过易于获得的小规模问题的全局最优调度方案训练得到. 最后, 通过与商业求解器Cplex以及现有算法的对比, 实验结果表明了所提算法的有效性.

英文摘要

      Refinery production scheduling is a mixed integer programming problem. As the scale increases, its solution time increases exponentially with the problem size, making it difficult to solve large-scale long-period oil refining production scheduling problems in a reasonable time. Aiming at this problem, this paper proposes a refinery production scheduling algorithm based on production task prediction and decomposition strategy, which can obtain a satisfactory solution to large-scale scheduling problems in a short time. The proposed algorithm decomposes the original problem into several single-period sub-problems with the same duration along the time axis, and designs a single-period production task prediction model based on deep learning to coordinate the solution of the sub-problems. The production task prediction model is trained through the data from small-scale problems, of which the global optimal solution is easy to obtain. Finally, by comparing with the commercial solver Cplex and existing algorithms, the experimental results show the effectiveness of the proposed algorithm.