| 融合深度反残差与注意力机制的山区高速公路事故严重程度预测模型 |
| |
| 引用本文： | 吕璞,柏强,陈琳.融合深度反残差与注意力机制的山区高速公路事故严重程度预测模型[J].中国公路学报,2021,34(6):205-213. |
| |
| 作者姓名： | 吕璞 柏强 陈琳 |
| |
| 作者单位： | 长安大学 运输工程学院, 陕西 西安 710064 |
| |
| 基金项目： | 国家自然科学基金项目（71901035）；陕西省自然科学基础研究计划项目（2019JM-228）；中央高校基本科研业务费专项资金项目（300102210113，300102218401） |
| |
| 摘 要： | 山区高速公路事故严重程度预测对保障交通安全具有重大意义。针对现有事故严重程度预测模型存在准确率低、泛化性差等问题，考虑到深度卷积神经网络可以高效处理图像问题，为此将事故影响因素图像化，提出一种融合深度反残差与注意力机制的山区高速公路事故严重程度预测模型。该模型首先采用相关性分析确定影响交通事故严重程度的因素，依据严重程度与影响因素将事故划分为财产损失、轻伤事故、重伤事故和死亡事故4类；然后将影响因素处理成图片的形式，进而将事故严重程度预测问题转化为图像的分类问题，随之构建基于反残差与注意力机制的山区高速公路事故严重程度预测模型，其中：基于深度可分离卷积的反残差结构可以以较少训练参数获取较高的准确率，基于软阈值的注意力机制作为一种非线性层可以忽略与事故严重程度无关的信息，Mish激活函数可以确保更好的信息流入神经网络。结果表明：在山区高速公路交通安全事故严重程度评估中，相比于传统的机器学习模型，所提出的模型识别准确率具有明显的提高，且测试准确率为85%左右，满足山区高速公路安全评估的实际预测需求。
|
| 关 键 词： | 交通工程 山区高速公路 反残差 注意力机制 神经网络 交通事故 |
| 收稿时间： | 2020-06-10 |
A Model for Predicting the Severity of Accidents on Mountainous Expressways Based on Deep Inverted Residuals and Attention Mechanisms |
| |
| LYU Pu,BAI Qiang,CHEN Lin.A Model for Predicting the Severity of Accidents on Mountainous Expressways Based on Deep Inverted Residuals and Attention Mechanisms[J].China Journal of Highway and Transport,2021,34(6):205-213. |
| |
| Authors: | LYU Pu BAI Qiang CHEN Lin |
| |
| Affiliation: | School of Transportation, Chang'an University, Xi'an 710064, Shaanxi, China |
| |
| Abstract: | The prediction of the severity of accidents on mountainous expressways is of great significance to ensure traffic safety. A model for predicting the severity of accidents on mountainous expressways based on deep inverted residuals and attention mechanisms is proposed to solve the problems of low accuracy and poor generalization in existing accident severity prediction models based on the accident influencing factors being visualized, inspired by the ideas that deep convolutional neural networks can efficiently handle image. First, the model uses correlation analysis to determine the factors that affect the severity of traffic accidents. According to the severity and influencing factors, the accident is divided into four categories: property loss, minor injury accidents, serious injury accidents, and fatal accidents. Second, these influencing factors are processed into the form of pictures, therefore, the problem of the prediction of the severity of accidents is transformed into the problem of image classification and a prediction model for the severity of accidents on mountainous expressways based on the inverted residual and attention mechanism is constructed. The inverted residuals based on depthwise convolution is designed to reduce the number of trainable parameters while maintaining high accuracy, the attention mechanism based on soft thresholding is developed as nonlinear transformation layers to effectively focus on feature related to driver behavior, and Mish is used as the activation function to allow better information to flow into the neural network.The results show that the model proposed in this paper has a significantly improved recognition accuracy, and the test accuracy rate is 85.21% in the evaluation of the severity of traffic safety accidents on mountainous expressways. Compared with traditional machine learning models, the proposed model meets the forecast demands of mountainous expressway safety evaluation. |
| |
| Keywords: | traffic engineering mountainous expressway inverted residual attention mechanism neural networks traffic accident |
| 本文献已被 CNKI 等数据库收录！ |
| 点击此处可从《中国公路学报》浏览原始摘要信息 |
|
点击此处可从《中国公路学报》下载全文 |
