[1]庄伟卿,余晗彧.考虑多头注意力机制的交通事故等级分类预测模型[J].福建理工大学学报,2025,23(06):599-510.[doi:10.3969/j.issn.2097-3853.2025.06.013]
　ZHUANG Weiqing,YU Hanyu.Traffic accident classification prediction model considering multi-head attention mechanism[J].Journal of Fujian University of Technology;,2025,23(06):599-510.[doi:10.3969/j.issn.2097-3853.2025.06.013]
点击复制

考虑多头注意力机制的交通事故等级分类预测模型()
分享到：

《福建理工大学学报》[ISSN:2097-3853/CN:35-1351/Z]

卷:
第23卷
期数:
2025年06期
页码:
599-510
栏目:
出版日期:
2025-12-25

文章信息/Info

Title:
Traffic accident classification prediction model considering multi-head attention mechanism
作者:
庄伟卿; 余晗彧
福建理工大学互联网经贸学院
Author(s):
ZHUANG Weiqing;  YU Hanyu
School of Internet Economics and Business, Fujian University of Technology
关键词:
道路交通; 交通事故等级分类; 风险预测; 集成模型; 机器学习
Keywords:
road traffic; traffic accident classification; risk prediction; ensemble model; machine learning
分类号:
U491.31
DOI:
10.3969/j.issn.2097-3853.2025.06.013
文献标志码:
A
摘要:
针对交通事故等级预测中多因素耦合导致的分类精度不足问题，本文提出一种考虑多头注意力机制的交通事故等级分类预测模型。该模型通过优化算法自动调参，结合卷积特征提取与时序依赖处理，并引入多头注意力机制强化关键特征学习。采用Kaggle平台上的公开数据集对模型进行训练与测试并与其他模型进行对比。实验结果表明，该模型在4 类事故的AUC 值分别达到0.984(1级)、0.989(2 级)、0.882(3 级)和0.803(4 级)，均为最高值。在分类难度最大的4 级事故预测中，该模型召回率表现显著领先于对比模型。在数据稀缺场景下，该模型的整体错误分类数最少且对2 级事故的预测仅出现1 例误判。
Abstract:
To address the challenge of inadequate classification accuracy in traffic accident severity prediction caused by multi-factor coupling effects, this study proposes a severity classification model considering multi-head attention mechanism. The model features an optimization algorithm for automated hyperparameter tuning, combines convolutional operations for feature extraction with sequential dependency modeling, and leverages multi-head attention to enhance the learning of discriminative features. Experiments were conducted on a publicly available Kaggle dataset, with comparative evaluations against baseline models. Key findings demonstrate that the model attains AUC values of 0.938 (Level 1), 0.989 (Level 2), 0.882 (Level 3), and 0.803 (Level 4) in four types of accidents, outperforming all competing methods. For the most challenging Level 4 accident prediction, this model achieves significantly superior recall performance compared with other models. Under data-scarce conditions, the proposed model exhibits remarkable robustness, yielding the lowest overall misclassification rate-including only one error for Level 2 predictions-among evaluated models.

参考文献/References:

［1］ 严利鑫, 胡鑫辉, 刘清梅, 等. 道路交通事故严重程度预测及致因分析 ［J］. 华东交通大学学报, 2024, 41（5）：65-73.［2］ ZHANG K，GUO Y M. Attention-based residual dilated network for traffic accident prediction［J］. Mathematics，2023，11（9）：2011.［3］ JIN Z X，NOH B. From prediction to prevention：leveraging deep learning in traffic accident prediction systems［J］. Electronics，2023，12（20）：4335.［4］ 于翔海，白佃国，于光，等. 基于XGBoost模型的城市道路实时交通事故风险预测研究［J］. 公路交通科技，2023，40（4）：237-247.［5］ 陈喜群，沈楼涛，李俊懿，等. 基于异构数据特征的城市轨道交通OD客流短时预测方法［J］. 交通信息与安全，2024，42（2）：158-165.［6］ EVEN S，ALBAYRAK A. Traffic accident severity prediction with ensemble learning methods［J］. Computers and Electrical Engineering，2024，114：109101.［7］ UGUZ S, BUYUKGOKOGLAN E. A hybrid CNNLSTM model for traffic accident frequency forecasting during the tourist season ［J］. Tehnicki VjesnikTechnical Gazette, 2022, 29(6): 2083-2089.［8］ XU X C，JIN X F，XIAO D Q，et al. A hybrid autoregressive fractionally integrated moving average and nonlinear autoregressive neural network model for short-term traffic flow prediction［J］. Journal of Intelligent Transportation Systems，2023，27（1）：1-18.［9］ VLAHOGIANNI E I，KARLAFTIS M G，GOLIAS J C. Optimized and meta-optimized neural networks for short-term traffic flow prediction：a genetic approach［J］. Transportation Research Part C：Emerging Technologies，2005，13（3）：211-234.［10］ QIU C Y，WANG C L，FANG B X，et al. Amultiobjective particle swarm optimizationbased partial classification for accident severity analysis［J］. Applied Artificial Intelligence，2014，28（6）：555-576.［11］ SOWMYA R，PREMKUMAR M，JANGIR P. Newtonraphsonbased optimizer：a new populationbased metaheuristic algorithm for continuous optimization problems［J］. Engineering Applications of Artificial Intelligence，2024，128：107532.［12］ CHENG S Y, LIU Y A. Research on transportation mode recognition based on multihead attention temporal convolutional network ［J］. Sensors, 2023, 23（7）：3585.

相似文献/References:

[1]贺肖,陈伯辉,沈斐敏.基于灰色马尔科夫链的道路交通死亡率预测[J].福建理工大学学报,2013,11(06):592.
[2]张扬永.基于TS-NN模型的道路交通车流量预测[J].福建理工大学学报,2021,19(06):560.[doi:10.3969/j.issn.1672-4348.2021.06.010]
　ZHANG Yangyong.Prediction of road traffic flow based on the TS-NN model[J].Journal of Fujian University of Technology;,2021,19(06):560.[doi:10.3969/j.issn.1672-4348.2021.06.010]

更新日期/Last Update: 2025-12-25