网络与信息安全学报 ›› 2020, Vol. 6 ›› Issue (6): 152-163.doi: 10.11959/j.issn.2096-109x.2020085

• 学术论文 • 上一篇    下一篇

基于循环神经网络的空载电动出租车的充电桩推荐方法

贾鉴(),刘林峰,吴家皋   

  1. 南京邮电大学计算机学院,江苏 南京 210023
  • 修回日期:2020-07-03 出版日期:2020-12-15 发布日期:2020-12-16
  • 作者简介:贾鉴(1996- ),女,山西长治人,南京邮电大学硕士生,主要研究方向为轨迹挖掘、车联网|刘林峰(1981- ),男,江苏丹阳人,博士,南京邮电大学教授,主要研究方向为移动计算、车联网、机器学习方法|吴家皋(1969- ),男,江苏苏州人,博士,南京邮电大学副教授,主要研究方向为容忍延迟网络、机会网络
  • 基金资助:
    国家自然科学基金(61872191)

Charging pile recommendation method for idle electric taxis based on recurrent neural network

Jian JIA(),Linfeng LIU,Jiagao WU   

  1. School of Computer Science,Nanjing University of Posts and Telecommunications,Nanjing 210023,China
  • Revised:2020-07-03 Online:2020-12-15 Published:2020-12-16
  • Supported by:
    The National Natural Science Foundation of China(61872191)

摘要:

提出了一种基于循环神经网络的空载电动出租车的充电桩推荐方法(CPRM-IET,charging pile recommendation method for idle electric taxis),来为空载状态下的电动出租车推荐最佳充电桩。空载状态下的电动出租车移动一般依赖于驾驶人的潜意识移动倾向和驾驶习惯,因此需要根据其历史移动轨迹来预测其未来移动,从而找到充电额外移动最小的若干充电桩。在CPRM-IET中,使用了一种基于双阶段注意力机制的循环神经网络(DA-RNN,dual-stage attention-based recurrent neural network)模型来预测电动出租车的未来轨迹,DA-RNN模型包括输入注意力机制和时间注意力机制。输入注意力机制在每个时刻为输入的行驶记录分配权重,而时间注意机制为编码器的隐藏状态分配权重。根据预测轨迹,再选择额外移动最小的若干充电桩,并推荐给电动出租车驾驶人。仿真结果表明,CPRM-IET可以在额外移动和均方根误差方面取得较好的结果,反映了CPRM-IET可以准确地预测空载电动出租车的未来轨迹,并向这些电动出租车推荐合适的充电桩。

关键词: 充电桩推荐, 循环神经网络, 输入注意力机制, 时间注意力机制, 轨迹预测

Abstract:

A charging pile recommendation method for idle electric taxis (CPRM-IET) based on recursive neural network was proposed to recommend the optimal charging piles for idle electric taxis.Usually,the movement of each idle electric taxi depends on the subconscious movement tendency and driving habits of the driver.Therefore,it is necessary to predict the future movement based on its historical movement trajectories,so as to find the charging piles with the least extra movements.In CPRM-IET,a dual-stage attention-based recurrent neural network (DA-RNN) model was provided to predict the future trajectories of electric taxis.DA-RNN model includes two types of attention mechanisms which are input attention mechanism and temporal attention mechanism.The input attention mechanism assigns different weights to the input driving sequence at each time slot,and the temporal attention mechanism assigns weights to the hidden state of the encoder.Based on the predicted future trajectories,several charging piles with the least extra movements were selected and recommended for the idle electric taxis.The simulation results show that CPRM-IET can achieve preferable results in terms of charging extra movement and root mean square error,which reflects that CPRM-IET can accurately predict the future trajectories of idle electric taxis and recommend optimal charging piles for these electric taxis.

Key words: charging pile recommendation, recurrent neural network, input attention mechanism, time attention mechanism, trajectory prediction

中图分类号: