| 基于表示学习的车辆到达时间预测 |
 点此下载全文 |
| 引用本文：李昕彤，李茂源，陈茜雅，司浩田.基于表示学习的车辆到达时间预测[J].计算技术与自动化,2024,(4):141-145 |
| 摘要点击次数: 138 |
| 全文下载次数: 0 |
|
|
| 中文摘要:随着智慧交通的兴起,人们对高效率出行的需求日益增加，因而寻求更优秀的模型来估计车辆行驶时间成为首要任务。由于交通系统有较强的非线性，并且受天气、时间等多种因素影响，综合考虑历史车流量信息和当前路况，提出了一种基于表示学习的多模态拟合模型。将车辆行驶时间估计(ETA)问题视为一个基于一组含有大量有效特征的纯时空序列的回归问题，分别采用不同的机器学习模型来解决每一部分的回归问题。通过滴滴出行的数据来训练模型，充分利用SDNE (Structure Deep Network Embedding)、LSTM (Long Short-Term Memory)、xDeepFM (eXtreme Deep Factorization Machine)算法的各自优势，最后的对比实验表明，提出的模型优于传统的深度学习算法。 |
| 中文关键词:智慧交通 表示学习 多模态拟合 机器学习 |
| |
| Prediction of Vehicle Reaching Time Based on Representation Learning |
|
|
| Abstract:With the rise of smart transportation, people’s demand for efficient travel is increasing.Therefore, the search for better models to estimate vehicle travel time has become a top priority. Since the traffic system has a strong nonlinearity and is affected by various factors such as weather and time, this paper proposes a multimodal fitting model based on representation learning, considering historical traffic flow information and current road conditions. This article regards the estimated time of arrival (ETA) problem as a regression problem based on a set of pure spatiotemporal sequences with a large number of effectivefeatures, and uses different machine learning models to solve each part of the regression problem. The model is trained through the data of Didi, and the respective advantages of SDNE(Structure Deep Network Embedding), LSTM(Long Short-Term Memory), xDeepFM(eXtreme Deep Factorization Machine) algorithms are fully utilized. Finally, the comparison test shows that the model proposed in this article is superior to the traditional deep learning algorithm. |
| keywords:smart transportation representation learning multi-modal fitting machine learning |
| 查看全文 查看/发表评论 下载pdf阅读器 |
位访问者,今日访问
