| 摘 要: 航班配餐备份量受到历史订票人数和多种外部因素的影响，可以利用注意力机制捕获航班历史订票人数对最终离港人数的影响程度，通过编解码器结构捕获外部复杂因素与航班旅客订票序列间的关系，再引入残差设计加快模型的收敛速度并防止梯度爆炸问题。文章提出一种基于注意力机制的航班配餐备份数预测算法，并在真实航班旅客数据集上进行实验，以MAPE(平均绝对百分比误差)为评价指标，该算法较传统的统计学方法、机器学习方法和深度学习方法，误差分别降低2.19%、0.42%和3.02%。 |
| 关键词: 航班配餐备份数；编解码器；注意力机制；残差设计；深度学习 |
|
中图分类号: TP39
文献标识码: A
|
| 基金项目: 厦门市重大科技项目资助(3502Z20201019). |
|
| An Algorithm for Predicting the Flight Catering Backups Quantity based on Attention Mechanism |
|
WANG Jun, WU Zixuan, DONG Jie, CAI Zheli, ZHANG Kai, XU Qiaoruo
|
(XIAMEN AIR, Xiamen 361000, China )
wangjun6@xiamenair.com; zixuan_wu@sina.com; dongjie3@xiamenair.com; caizheli@xiamenair.com; zhangkai7@xiamenair.com; xuqiaoruo@xiamenair.com
|
| Abstract: The backups quantity of flight catering is affected by the number of historical bookings and various external factors. The attention mechanism can be used to capture the different influence grade of the number of flight history bookings on the final departures. The Encoder-Decoder structure can capture the relationship between the external complex factors and the flight passenger booking sequence. The residual design is introduced to accelerate the convergence speed of the model and prevent the gradient explosion problem. This paper proposes an algorithm for predicting the flight catering backups quantity based on attention mechanism. Taking MAPE (Mean Absolute Percentage Error) as the evaluation index, the empirical studies on real flight passenger datasets show that compared with the traditional statistical methods, machine learning methods and deep learning methods, the proposed algorithm reduces the error by 2.19%, 0.42% and 3.02%, respectively. |
| Keywords: flight catering backups quantity; encoder-decoder; attention mechanism; residual design; deep learning |
