深度时空残差网络预测城市人群流动

需积分: 0 157 浏览量更新于2024-08-05 收藏 2.38MB PDF 举报

"这篇论文是'深度时空残差网络用于城市范围内的客流预测'，由Zhang等人在AAAI 2017会议上发表。它介绍了一种名为ST-ResNet的深度学习方法，用于预测城市的区域间客流流入和流出，以支持交通管理和公共安全。该方法针对复杂因素如地区间交通、事件和天气对客流的影响，设计了一个端到端的ST-ResNet结构。" 正文：在现代城市环境中，准确预测客流流量对于交通管理以及公共安全具有重要意义。然而，由于客流受多种复杂因素（如跨区域交通动态、社会活动和天气变化）的影响，这一任务极具挑战性。Zhang等人的研究工作提出了一个创新的解决方案——ST-ResNet（深度时空残差网络），这是一种专为处理时空数据特性设计的深度学习模型。 ST-ResNet的核心是利用了残差神经网络（Residual Neural Network）框架来捕捉时空数据中的关键属性。残差网络在图像识别领域已经证明了其在解决深层网络梯度消失问题上的优势，通过短接路径使得信息可以直接传递，有助于训练更深的网络结构。在ST-ResNet中，这种框架被用来模拟客流流量的时间连续性、周期性和趋势性。时间连续性是客流流量的一个基本特征，ST-FlowNet通过残差卷积单元来捕捉这一特性，这使得模型能够学习到随着时间推移的流量变化模式。周期性体现在人们日常活动的规律性，例如工作日与周末的流量差异，早晚高峰时段的流量变化等。ST-ResNet通过专门设计的分支来捕捉这些周期性的模式，增强了预测的准确性。趋势性则反映了长期的流量变化，可能由城市发展、新设施的建设和公众行为模式的改变引起。ST-ResNet的残差网络结构使得模型能够学习到并预测这些长期的演变趋势。此外，ST-ResNet作为一个端到端的系统，能够同时预测城市中所有区域的流入和流出流量，这在实际应用中非常实用，因为交通流量往往是相互关联的，一个区域的流量变化会影响到其他区域。通过整合空间依赖性，ST-ResNet可以更全面地理解和预测城市交通的整体动态。论文“ST-ResNet-AAAI17-Zhang1”提出的ST-ResNet模型为复杂的城市客流预测提供了一个强大而有效的工具，它融合了深度学习和时空数据的特性，有望改善交通管理和公共服务的效率。这一方法对于理解和预测大规模城市系统的复杂动态具有深远的理论与实践意义。

Deep Spatio-Temporal Residual Networks for Citywide Crowd Flows Prediction

∗

Junbo Zhang

, Yu Zheng

1,2,3,4†

, Dekang Qi

2,1

Microsoft Research, Beijing, China

School of Information Science and Technology, Southwest Jiaotong University, Chengdu, China

School of Computer Science and Technology, Xidian University, China

Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences

{junbo.zhang, yuzheng}@microsoft.com, dekangqi@outlook.com

Abstract

Forecasting the ﬂow of crowds is of great importance to trafﬁc

management and public safety, and very challenging as it is

affected by many complex factors, such as inter-region traf-

ﬁc, events, and weather. We propose a deep-learning-based

approach, called ST-ResNet, to collectively forecast the in-

ﬂow and outﬂow of crowds in each and every region of a

city. We design an end-to-end structure of ST-ResNet based

on unique properties of spatio-temporal data. More speciﬁ-

cally, we employ the residual neural network framework to

model the temporal closeness, period, and trend properties

of crowd trafﬁc. For each property, we design a branch of

residual convolutional units, each of which models the spa-

tial properties of crowd trafﬁc. ST-ResNet learns to dynam-

ically aggregate the output of the three residual neural net-

works based on data, assigning different weights to different

branches and regions. The aggregation is further combined

with external factors, such as weather and day of the week,

to predict the ﬁnal trafﬁc of crowds in each and every region.

Experiments on two types of crowd ﬂows in Beijing and New

York City (NYC) demonstrate that the proposed ST-ResNet

outperforms six well-known methods.

Introduction

Predicting crowd ﬂows in a city is of great importance to

trafﬁc management and public safety (Zheng et al. 2014).

For instance, massive crowds of people streamed into a strip

region at the 2015 New Year’s Eve celebrations in Shanghai,

resulting in a catastrophic stampede that killed 36 people. In

mid-July of 2016, hundreds of “Pokemon Go” players ran

through New York City’s Central Park in hopes of catching

a particularly rare digital monster, leading to a dangerous

stampede there. If one can predict the crowd ﬂow in a re-

gion, such tragedies can be mitigated or prevented by utiliz-

ing emergency mechanisms, such as conducting trafﬁc con-

trol, sending out warnings, or evacuating people, in advance.

In this paper, we predict two types of crowd ﬂows (Zhang

et al. 2016): inﬂow and outﬂow, as shown in Figure 1(a).

Inﬂow is the total trafﬁc of crowds entering a region from

∗

This research was supported by NSFC (Nos. 61672399,

U1401258), and the 973 Program (No. 2015CB352400).

†

Correspondence author. This work was done when the third

author was an intern at Microsoft Research.

 2017, Association for the Advancement of Artiﬁcial

other places during a given time interval. Outﬂow denotes

the total trafﬁc of crowds leaving a region for other places

during a given time interval. Both ﬂows track the transition

of crowds between regions. Knowing them is very beneﬁcial

for risk assessment and trafﬁc management. Inﬂow/outﬂow

can be measured by the number of pedestrians, the number

of cars driven nearby roads, the number of people traveling

on public transportation systems (e.g., metro, bus), or all of

them together if data is available. Figure 1(b) presents an

example. We can use mobile phone signals to measure the

number of pedestrians, showing that the inﬂow and outﬂow

of r

are (3, 1) respectively. Similarly, using the GPS trajec-

tories of vehicles, two types of ﬂows are (0, 3) respectively.

(a) Inﬂow and outﬂow (b) Measurement of ﬂows

Figure 1: Crowd ﬂows in a region

Simultaneously forecasting the inﬂow and outﬂow of

crowds in each region of a city, however, is very challenging,

affected by the following three complex factors:

1. Spatial dependencies. The inﬂow of Region r

(shown in

Figure 1(a)) is affected by outﬂows of nearby regions (like

) as well as distant regions. Likewise, the outﬂow of r

would affect inﬂows of other regions (e.g., r

). The inﬂow

of region r

would affect its own outﬂow as well.

2. Temporal dependencies. The ﬂow of crowds in a region

is affected by recent time intervals, both near and far. For

instance, a trafﬁc congestion occurring at 8am will affect

that of 9am. In addition, trafﬁc conditions during morning

rush hours may be similar on consecutive workdays, re-

peating every 24 hours. Furthermore, morning rush hours

may gradually happen later as winter comes. When the

temperature gradually drops and the sun rises later in the

day, people get up later and later.

3. External inﬂuence. Some external factors, such as weather

conditions and events may change the ﬂow of crowds

tremendously in different regions of a city.

下载后可阅读完整内容，剩余6页未读，立即下载

Period熹微

粉丝: 30
资源: 307

深度时空残差网络预测城市人群流动

ResNet论文资料共3个文档 1-原版论文pdf-2-中文翻译pdf-3-中英文翻译对照pdf.rar

场景分类places365

LeNet+AlexNet+ResNet论文

ST-ResNet训练算法

STResNet:深度时空残留网络（ST-ResNet）的TensorFlow实现

PyTorch-SE-ResNet:SE-ResNet PyTorch版本

数据融合matlab代码-AMDF-ResNet:AMDF-ResNet

matlab调整代码大小写-st-resnet:NIPS16和“用于视频动作识别的时空乘法器网络”，CVPR17的“用于视频动作识别的时空残留

keras-inception-resnet-v2:使用Keras的Inception-ResNet v2模型（带有权重文件）

SENet.mxnet:挤压和激励网络（SE-ResNext，SE-Resnet，SE-Inception-v4和SE-Inception-Resnet-v2）的MXNet实现。

最新资源