计算机与现代化 ›› 2021, Vol. 0 ›› Issue (02): 56-61.

• 人工智能 • 上一篇    下一篇

基于深度网络的出租车Pick-up需求预测

  

  1. (杭州电子科技大学计算机学院,浙江杭州310018)
  • 出版日期:2021-03-01 发布日期:2021-03-01
  • 作者简介:李巍(1994—),男,湖北鄂州人,硕士研究生,研究方向:数据挖掘,E-mail: li_weihh@163.com。
  • 基金资助:
    工信部产业技术基础公共服务平台项目(2019-00899-3-1)

Taxi Pick-up Demand Prediction Based on Deep Networks for

  1. (School of Computer Science and Technology, Hangzhou Dianzi University, Hangzhou 310018, China)
  • Online:2021-03-01 Published:2021-03-01

摘要: 在智慧城市建设中,区域的出租车需求预测是一个十分重要的模块。为了预测指定区域未来时刻的出租车需求,本文通过拓展已有序列模型,提出一个多时间分辨率的基于层次注意力机制的循环高速网络(Multi-Time Resolution Hierarchical Attention-Based Recurrent Highway Networks, MTR-HRHN)。MTR-HRHN将对外生数据时空特征的提取和目标数据的时空建模整合到单个框架中,并且通过多分辨率(例如每个小时或者每天)对序列数据不同的时间属性建模,从而捕获更全面的时间模式。最后,在纽约市出租车数据集上评估MTR-HRHN的预测性能。实验结果表明,与其他经典时间序列预测方法相比,MTR-HRHN在多个高需求区域的短期需求预测上表现出更好的预测性能。

关键词: 需求预测, 深度学习, 注意力机制, 时间分辨率

Abstract: The taxi demand prediction is an important part in the construction of smart cities. In order to predict the taxi demand in the designated regions in the future, this paper proposes a multi-time resolution hierarchical attention-based recurrent highway networks (MTR-HRHN) through expanding existing prediction models. The MTR-HRHN integrates the extraction of spatiotemporal features of exogenous data and the spatiotemporal modeling of target data into a single framework, builds a model based on the different temporal properties of sequential data through multiple resolutions (such as every hour or every day), so as to capture a more comprehensive time pattern. Finally, this paper evaluates the prediction performance of MTR-HRHN on the New York City taxi dataset. The experimental results show that compared with other classical time series prediction methods, MTR-HRHN has better prediction performance in short-term demand prediction in multiple high-demand regions.


Key words: demand prediction, deep learning, attention mechanism, time resolution