«上一篇/Previous Article|本期目录/Table of Contents|下一篇/Next Article»

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

33

期数:

2023年02期

页码:

195-202

栏目:

人工智能

出版日期:

2023-02-10

文章信息/Info

Title:

Approach of Predicting Number of Citywide Traffic Accidents Using Long Short-term Memory Neural Network

文章编号:

1673-629X(2023)02-0195-08

作者:

孙振华¹; 2 ; 王转转² ; 肖 鑫²

1. 绍兴市交通建设有限公司,浙江 绍兴 321000;

2. 长安大学 信息工程学院,陕西 西安 710064

Author(s):

SUN Zhen-hua1; 2 ; WANG Zhuan-zhuan2 ; XIAO Xin2

1. Shaoxing Transportation Construction Co. ,Ltd. ,Shaoxing 321000,China;

2. School of Information Engineering,Chang’an University,Xi’an 710064,China

关键词:

交通事故; 神经网络; 长短期记忆; 时间序列; 最优窗口

Keywords:

traffic accidents; neural network; LSTM; time series; optimum window

分类号:

TP399

DOI:

10. 3969 / j. issn. 1673-629X. 2023. 02. 029

摘要:

市级交通事故量时间序列的波动是影响对其准确预测的关键因素。 提出的预测方法针对市级日交通事故量时间序列,采用长短期记忆神经网络( Long Short-Term Memory,LSTM)捕捉序列当前观测值与前序观测值的时序依赖关系,通过寻找最优窗口长度的 LSTM 市级日粒度交通事故量预测模型使拟合数据对训练集误差最小,对验证集的预测结果在转为周粒度时取得了较为准确的预测效果。 提出的预测方法解决了影响市级周交通事故量准确预测的问题,该方法发现基于交通事故量训练的用于捕获观测值时序依赖关系的 LSTM 模型对数据基本趋势准确性的表达远好于对数据波动性的表达。 为此,提出最优窗口算法来确定 LSTM 模型最优窗口长度,以确保对训练集基本趋势表达的准确性,再根据所发现的细粒度下的预测结果对交通事故量基本趋势的准确描述可转化为粗粒度下对波动性准确描述的事实,将日粒度预测结果转为周粒度后就取得了较为准确的预测效果。

Abstract:

The fluctuation of the number of citywide traffic accidents is a key to affect the accuracy of prediction. The proposedapproach,in the time-series observations at the day level coming from the statistics of urban traffic accidents,adopted the Long Short-Term Memory ( LSTM) to capture the dependent relationship between the current observations and the preceding observations. TheLSTM model for predicting the number of citywide traffic accidents per day corresponding to the optimum window length of the input sequences at the day level was developed to achieve the minimum error between the fitting data and the observations of the training set. Asa result,the fluctuation of the validation set at week level can be depicted accurately when the predicted data are aggregated at week level.The key point to compromise the accuracy of prediction has been addressed in the proposed approach,in which,the LSTM model forcapturing the dependent relationships between observations,trained by the number of citywide traffic accidents,is found out to perform thedescription of the trend far more accurate than the fluctuation of data. Then an algorithm was proposed to determine the optimum windowlength for the LSTM model to ensure the accuracy of depicting the trend of the training set. According to the fact that the predicted dataperform the accurate description of the trend of the number of citywide traffic accidents in fine-grained time intervals will result in the accurate description of the fluctuation in coarse - grained time intervals, the more accurate prediction effects have been gained when thepredicted data at the day level are aggregated at the week level.

常用功能

更新日期/Last Update: 2023-02-10