中文核心期刊
CSCD来源期刊
中国科技核心期刊
RCCSE中国核心学术期刊

重庆交通大学学报(自然科学版) ›› 2024, Vol. 43 ›› Issue (12): 92-97.DOI: 10.3969/j.issn.1674-0696.2024.12.12

• 交通+大数据人工智能 • 上一篇    

基于多尺度混合注意力机制卷积神经网络的伦理决策模型设计

刘国满1,2,罗玉峰2, 盛敬1,陶珍1   

  1. (1. 南昌工程学院 机械工程学院,江西 南昌330099; 2. 华东交通大学 电气与自动化工程学院,江西 南昌330013)
  • 收稿日期:2024-06-26 修回日期:2024-10-08 发布日期:2024-12-24
  • 作者简介:刘国满(1975—),男,安徽枞阳人,讲师,博士,主要从事无人驾驶伦理决策方面的研究。E-mail:jerry@nit.edu.cn
  • 基金资助:
    江西省教育厅自然科学基金项目(GJJ201911)

Design of Ethical Decision-Making Model Based on Multi-scale Attention Mechanism Convolutional Neural Network

LIU Guoman1,2, LUO Yufeng2, SHENG Jing1, TAO Zhen1   

  1. (1. School of Mechanical Engineering, Nanchang Institute of Technology, Nanchang 330099, Jiangxi, China; 2. School of Electrical and Automation Engineering, East China Jiaotong University, Nanchang 330013, Jiangxi, China)
  • Received:2024-06-26 Revised:2024-10-08 Published:2024-12-24

摘要: 针对目前自动驾驶汽车伦理困境下很难做出确定、合理的决策,阻碍了自动驾驶技术的问题,设计了一种基于多尺度混合注意力机制卷积神经网络的自动驾驶汽车伦理决策模型。首先,依据卷积神经网络结构和功能以及伦理困境特点,设计卷积神经网络训练模型和参数,运用训练集中伦理困境对该训练模型进行训练,构建多尺度混合注意力机制卷积神经网络模型;然后,运用测试集中伦理困境对该卷积神经网络模型进行测试和验证。结果表明:多尺度卷积神经网络模型相对于传统卷积神经网络,准确率方面有了较大改进,加入注意力机制卷积神经网络模型相对于未加入模型,稳定性得到加强;多尺度混合注意力机制卷积神经网络模型的准确率和稳定性都较高,最高准确率达到了89%。

关键词: 交通工程;自动驾驶汽车;伦理决策;注意力机制;卷积神经网络

Abstract: In order to solve the problem that it is difficult to make certain and reasonable decisions under the current ethical dilemma of autonomous vehicle, which hinders the application and development of self-driving technology, therefore, an ethical decision-making model based on multi-scale mixed attention mechanism convolutional neural network (CNN) was designed for autonomous vehicles. Firstly, according to the structure and function of CNN as well as the characteristics of ethical dilemmas, the training model and parameters of CNN were designed, then ethical dilemmas in training set were used to train the proposed training model, and the model based on multi-scale mixed attention mechanism CNN was constructed. Secondly, the ethical dilemmas in testing set were used to test and verify the proposed CNN model. The results show that multi-scale CNN model has better accuracy than traditional CNN does. Furthermore, the stability of CNN model with attention mechanism is enhanced compared to the model without attention mechanism. Therefore, the multi-scale mixed attention mechanism CNN model has higher accuracy and stability, with the highest accuracy rate reaching 89%.

Key words: traffic engineering; autonomous vehicles; ethical decision-making; attention mechanism; CNN

中图分类号: