基于时空的深度学习模型感知通行时间  

Spatial-Temporal Based Deep Learning Model Perceives Travel Time

作　　者：刘阳[1] 李建波[1] 楚本嘉 马照斌 夏丰千 

机构地区：[1]青岛大学城市计算机科学技术学院,山东 青岛 [2]青岛大学泛在网络与城市计算研究所,山东 青岛

出　　处：《计算机科学与应用》2023年第3期378-389,共12页Computer Science and Application

摘　　要：随着国民经济的快速增长,人们的生活水平日益提高,私家车的数量不断增加,导致城市出现一系列交通拥堵问题和事故。因此,对于城市交通监控、导航、路线规划和乘车共享来说,感知给定城市路径的通行时间至关重要。以前的方法总是感知单个路径的通行时间,然后将它们相加为整个路径的通行时间。我们提出了一个基于时空的深度学习框架来感知整个路径的通行时间。更具体地说,我们使用卷积神经网络来捕获时间和空间依赖性。由于还有一些影响因素(如天气、时间、驾驶员等)影响通行时间,我们添加了一个影响因素模块来预处理数据。大量的实验证明,我们提出的模型显著优于其他已知模型。With the rapid growth of the national economy, people’s living standards are improving day by day, and the number of private cars is increasing, leading to a series of traffic congestion problems and accidents in cities. Therefore, it is essential for urban traffic monitoring, navigation, route planning, and ride sharing to perceive the travel time of a given urban path. Previous methods always perceived the travel time of a single path, and then summed them as the travel time of the whole path. We propose a spatial-temporal-based deep learning framework to perceive the travel time of the entire path. More specifically, we use this convolutional neural network to capture both temporal and spatial dependencies. Since there are also some impact factors (such as weather, time, driver, etc.) that affect the travel time, we have added an impact factor module to preprocess the data. Extensive experiments have shown that our proposed model is significantly outperforming other known models.

关 键 词：通行时间感知 深度学习 时间–空间 

分 类 号：G63[文化科学—教育学]