基于卷积神经网络的机票低价预测  

Air ticket low-price prediction based on convolution neural network

作　　者：林友芳[1,2] 蒋鹏 郭晟楠 武志昊 LIN Youfang;JIANG Peng;GUO Shengnan;WU Zhihao(School of Computer and Information Technology,Beijing Jiaotong University,Beijing 100044,China;CAAC Key Lab of Intelligent Passenger Service of Civil Aviation,Beijing 101318,China)

机构地区：[1]北京交通大学计算机与信息技术学院,北京100044 [2]民航旅客服务智能化应用技术重点实验室,北京101318

出　　处：《北京交通大学学报》2019年第5期1-9,共9页JOURNAL OF BEIJING JIAOTONG UNIVERSITY

基　　金：中央高校基本科研业务费专项资金(2019JBM023);国家自然科学基金(61603028)~~

摘　　要：准确的机票低价预测有助于民航需求与供给的灵活对接及民航资源的充分利用.机票价格波动性大、随机性强、易受到诸多因素的影响,使得机票价格预测成为了一个极具挑战的问题.充分考虑机票价格自身特点,设计了二维"机票价格时间片"结构,并基于时间片充分挖掘、利用机票价格数据的规律与关系,设计了以卷积神经网络为核心的两阶段机票价格预测模型,对未来机票最低价格进行预测.在某在线订票网站的真实价格数据集上进行了验证,并与4种流行的基准模型进行了对比.结果表明:本文的方法明显优于其他模型,MAE效果提升了13.67%,MAPE数值降低了1.52%.Accurate low-price air ticket forecasting facilitates the flexible docking of civil aviation demand and supply.It also helps to achieve full utilization of civil aviation resources.Fluctuations of ticket price are large,random,and vulnerable to many factors,making ticket price forecasting a challenging issue.This paper fully considers the characteristics of the ticket price,and designs a two-dimensional"air ticket price time slice"structure.Based on the time slice,the law and relationship of ticket price data can be futher exploited.Besides,the core two-stage ticket price forecasting model is proposed to predict the lowest price of future tickets based on convolutional neural network.Experiments are conducted on the real price dataset of an online booking website and compared with four popular benchmark models.The results show that the proposed method is obviously better than other models.MAE and MAPE achieve improvements of 13.67% and 1.52%respectively.

关 键 词：深度学习 机票低价预测 卷积神经网络 价格序列 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]