基于路网结构的城市交通事故短期风险预测方法

PDF下载HTML

Title:

A short-term risk prediction method for urban traffic accidents based on road network

作者:

张延孔¹, 卢家品¹, 张帅超², 姬晓鹏², 罗月童¹, 陈为²

1. 合肥工业大学 计算机与信息学院，安徽 合肥 230000;
2. 浙江大学 计算机科学与技术学院，浙江 杭州 310018

Author(s):

ZHANG Yankong¹, LU Jiapin¹, ZHANG Shuaichao², JI Xiaopeng², LUO Yuetong¹, CHEN Wei²

1. School of Computer Science and Information Engineering, Hefei University of Technology, Hefei 230000, China;
2. College of Computer Science and Technology, Zhejiang University, Hangzhou 310018, China

关键词:

图卷积; 交通事故; 事故模式; 多源数据; 风险预测; 路网结构; 长短期记忆网络; 智慧城市

Keywords:

GCNN; traffic accident; accident mode; multi-source data; risk forecasting; road network structure; LSTM; smart city

分类号:

TP18

DOI:

10.11992/tis.201910002

摘要:

城市交通事故一般都发生在公共道路上，然而现有的交通事故风险预测算法都通过对预测区域进行规则网格化来确定预测空间单位，导致预测精度不高且实用价值较低。本文将道路路段作为预测单位，采用图卷积和长短期记忆网络，构建了一种基于路网结构的城市交通事故短期风险预测方法(traffic accidents risk prediction based on road network，TARPBRN)。该方法能对指定路段短期内的交通事故风险进行预测，从而可以有针对性地进行治理，减少交通事故的发生。本文使用杭州市西湖区的交通事故数据对模型进行了训练，并与4种常用的计量经济学模型和3种已有的深度学习预测算法进行了对比。实验结果证明本文算法在准确度、正确率和漏报率等方面都优于已有算法。

Abstract:

Urban traffic accidents usually occur on public roads. However, the existing traffic accident risk prediction algorithms determine the prediction space unit by regularizing grid of the prediction area, which leads to low prediction accuracy and low practicability. Taking road sections as the prediction unit, this paper constructs a short-term traffic accident risk prediction method based on road network structure (TARPBRN) by using graph convolution and long short-term memory network. This method can predict the traffic accident risk in a short period of the designated section, so as to carry out targeted governance and reduce the occurrence of traffic accidents. In this paper, traffic accident data from Xihu District, Hangzhou city are used to train the model, and four econometric models and three existing deep learning prediction algorithms are compared. The experimental results show that the proposed algorithm is superior to the existing ones in accuracy, precision and false negative rate (FNR).