基于模仿学习的机场停机位再分配决策算法

针对机位再分配算法结果难以满足不同操作人员操作习惯的问题，提出一种符合实际业务人员操作习惯的机位再分配推荐算法。首先以航班特征属性和停机位的资源占用状态构建决策环境空间模型，将人工操作数据转换为多通道时空矩阵，再以卷积神经网络构建的生成对抗网络（generative adversarial network，GAN）拟合其序贯决策操作策略。仿真结果表明，可靠度在90%以上的调整动作占比最高达到84.4%。经过在三个数据集上的测试，模型对不同来源的操作数据具有较好的区分能力。对比不同扰动下的动态调整结果，算法能够得到航班—机位属性特征与原有人工操作属性特征接近的调整方案。

Decision-making algorithm for airport gate reassignment based on imitation learning

In order to solve the problem that the results of the gate reassignment algorithm can''t meet habits of different operators, this paper proposed a method that accordsed with the actual operators'' operating habits. Firstly, this paper established the spatial model of decision-making environment by using flights characteristics and occupancy of gate resources. The model transformed manual operating data into a multi-channel time-space matrix. Then, it made use of CNN-based generative adversarial network to match the order decision-making operation strategy. The simulation results show that actions with reliability scores of more than 90% account for up to 84.4%. The model has a good ability to distinguish the operation data from 3 different operators. Compared with dynamic adjustment result under perturbance, this algorithm can obtain an adjustment scheme whose flight-gate attribute characteristics are closer to the original manual operation.