摘要
准确高效识别旅客群体是铁路制定运营策略以及提高旅客服务质量的关键因素之一。以铁路旅客实名制购票数据以及出行数据为基础,构建旅客个体特征以及社交关系特征特性,并进行向量化表示,综合考虑现有准确度较优的深度学习模型存在的模型复杂、计算量大等问题,利用知识蒸馏以及边缘计算相结合的思路,基于知识蒸馏实现模型压缩降低内存成本和加速计算,同时对训练的模型进行分割,采用边缘计算就近实时计算的特性,将拆分后的部分模型部署在应用服务器节点,从而减低了模型的推理时延与能耗,以适应实时售票过程中对旅客群体的实时划分。最后以冰雪快运服务推荐为例,验证该模型的准确率以及预测效率,为铁路针对旅客群体的个性化客运产品设计以及营销策略优化提供决策建议。
Accurate and efficient identification of passenger groups is one of the key factors in formulating operation strategies and improving passenger service quality.Based on the real-name ticket purchase data and travel data of railway passengers,the individual characteristics and social relationship characteristics of passengers were constructed and vectorized.Taking into overall consideration the complexity and heavy calculation workload of the existing deep learning model with better accuracy,this paper,combining knowledge distillation and edge calculation,achieved model compression based on knowledge distillation to reduce memory cost and accelerate calculation.At the same time,segmenting the training model and adopting the nearest real-time computing characteristics of edge computing,it deployed part of the split model in the application server node,thus reducing the reasoning delay and energy consumption of the model,so as to adapt to the real-time division of the passenger group in real-time ticketing.Finally,with ice and snow express service recommendation as an example,it verified the accuracy and prediction efficiency of the model,and provided decision-making suggestions on personalized passenger transport product design and optimized marketing strategy of the railway for passenger groups.
作者
梅巧玲
郝晓培
杨立鹏
易超
MEI Qiaoling;HAO Xiaopei;YANG Lipeng;YI Chao(Institute of Computing Technology,China Academy of Railway Sciences Corporation Limited,Beijing 10081,China)
出处
《铁道运输与经济》
北大核心
2024年第4期27-33,67,共8页
Railway Transport and Economy
基金
国家重点研发计划项目(2020YFF0304101)。
关键词
产品设计
个体特征
社交关系
群体划分
边缘计算
知识蒸馏
Product Design
Individual Characteristics
Social Relationship
Group Division
Edge Calculation
Knowledge Distillation
