The personalized crew rostering problem (CRP) consists of assigning pairings (sequences of flights, deadheads, connections, and rests, forming one or several days of work) to individual crew members to create a feasible roster that maximizes crew satisfaction. This problem is often solved using a branch-and-price algorithm. In this paper, we propose a partial pricing scheme for the CRP in which the column generation subproblem of each crew member only contains the pairings that are likely to be selected in an optimal or near-optimal solution. The task of selecting which pairings to include in each network is performed by a deep neural network trained on historical data. We test the proposed method on several large instances. Our results show that our method finds solutions of similar quality as that of the classical branch-and-price algorithm in less than half of the computational time.

个性化机组排班问题 (CRP) 包括为单个机组成员分配配对（飞行顺序、空头、连接和休息，形成一天或几天的工作），以创建一个可行的名单，最大限度地提高机组人员的满意度。这个问题通常使用分支和价格算法来解决。在本文中，我们提出了 CRP 的部分定价方案，其中每个机组成员的列生成子问题仅包含可能在最优或接近最优解中选择的配对。选择要包含在每个网络中的配对的任务由经过历史数据训练的深度神经网络执行。我们在几个大型实例上测试了所提出的方法。