交通运输系统工程与信息 ›› 2019, Vol. 19 ›› Issue (4): 108-114.

• 系统工程理论与方法 • 上一篇    下一篇

基于RBM-GASA-BPNN 的潜在高价值旅客预测

徐涛* 1, 2, 3,刘泽君1,卢敏1, 2, 3   

  1. 1. 中国民航大学计算机科学与技术学院,天津 300300;2. 中国民航信息技术科研基地,天津 300300; 3. 民航旅客服务智能化应用技术重点实验室,北京 101318
  • 收稿日期:2019-01-17 修回日期:2019-03-16 出版日期:2019-08-25 发布日期:2019-08-26
  • 作者简介:徐涛(1962-),男,重庆人,教授.
  • 基金资助:

    国家自然科学基金/National Natural Science Foundation of China(61502499);中国民航大学科研基金/The Scientific Research Foundation from Civil Aviation University of China (2013QD18X);民航旅客服务智能化应用技术重点实验室项目/ Key Laboratory of Intelligent Passenger Service of Civil Aviation, CAAC (TS-CAKL-2018-01).

Potential High Value Passenger Forecast Based on RBM-GASA-BPNN

XU Tao1, 2, 3, LIU Ze-jun1, LU Min1, 2, 3   

  1. 1. College of Computer Science and Technology, Civil Aviation University of China, Tianjin 300300, China; 2. Information Technology Research Base of Civil Aviation Administration of China, Tianjin 300300, China; 3. Key Laboratory of Intelligent Passenger Service of Civil Aviation, CAAC, Beijing 101318, China
  • Received:2019-01-17 Revised:2019-03-16 Online:2019-08-25 Published:2019-08-26

摘要:

针对用BP神经网络(Back Propagation Neural Network, BPNN)进行潜在高价值旅客预测时出现的特征表达能力弱、稳定性差、易陷入局部极值的不足,提出一种新颖的基于 RBM-GASA-BPNN的潜在高价值旅客预测方法.该方法首先通过聚类算法划分旅客类别,设置类别标签;然后利用受限玻尔兹曼机(Restricted Boltzmann Machine, RBM)提取旅客行为特征并确定最优BPNN 初始权值和偏置的寻优范围,又利用遗传模拟退火算法(Genetic Algorithm-Simulate Anneal, GASA)对BPNN参数进行精调,确定了最优的BPNN初始权值和偏置;最后,利用优化后的BPNN对旅客进行分类预测.实验结果表明,本文提出的方法克服了基于BPNN的分类预测方法的缺陷,具有更高的分类预测准确率和潜在高价值旅客预测能力.

关键词: 航空运输, 价值类别预测, 预测模型, 潜在高价值旅客, RBM, GASA, BPNN

Abstract:

Aiming at the problems of weak feature expression ability, poor stability and higher local optimal probability of the potential high-value passengers discovery method based on BP neural network (BPNN), a novel potential high value passengers discovery method based on RBM-GASA-BPNN is proposed in this paper. Firstly, clustering algorithm is used to classify passengers and set category labels. Then the restricted Boltzmann machine (RBM) is used to automatically extract the passenger's behavior features and provide the optimal range of initial weight and bias for BPNN. And the genetic algorithm-simulated annealing (GASA) algorithm is used to adjust the parameters precisely to find the optimal initial weight and bias of BPNN. Finally, the optimized BPNN is used to classify passengers. The experimental results show that the proposed method overcomes the shortcomings of existing method based on BPNN and has a better classification prediction accuracy and potential high value passenger forecast ability.

Key words: air transportation, value category prediction, prediction model, potential high value passenger, RBM, GASA, BPNN

中图分类号: