基于ASPC-DA的假冒绿通车识别方法
Identification method of Fake Toll-free Vehicles Based on ASPC-DA Model
徐旭东 1易洪波 2何伊伦1
作者信息
- 1. 四川省交通运输发展战略和规划科学研究院,成都 610000
- 2. 西南交通大学,成都 611756
- 折叠
摘要
"绿色通道"政策的提出是为构建高效农产品流通网络,但存在不法车辆伪装成"绿通车"通行并逃缴高速通行费用的情况.针对该问题,提出了一种基于ASPC-DA的假冒绿通车识别方法.选取四川省正常绿通车通行历史数据以及假冒绿通车通行历史数据作为训练模型的案例数据集,并进行数据预处理及数据增强操作.然后基于PyTorch深度学习框架进行ASPC-DA-Ⅰ模型预训练阶段,预训练得到的模型具有较高的识别率,但稍逊于现有文献方法.在ASPC-DA-Ⅱ模型微调阶段,引入自步学习对模型进行逐步调整最终得到ASPC-DA模型,识别率为92.69％,远优于现有方法.通过灵敏度分析结果表明,ASPC-DA模型最优训练样本量为总样本量的70％,最优迭代上限为90～100,最优终止条件阈值为10-5～10-4.
Abstract
To build an efficient circulation network for agricultural goods,the"green channel"policy was proposed.However,there are some vehicles disguised as"green vehicles"try to passing through the highway to avoid paying the toll.In this paper,we consider historical data of normal toll-free vehicle data and fake toll-free vehicle data as case data set.Data pre-processing and aug-mentation are conduct at first.Then we pre-train the model in stage ASPC-DA-Ⅰ based on the PyTorch deep learning framework,the recognition rate of which is acceptable.Thirdly,in the ASPC-DA-Ⅱ stage,we introduce self-pace learning to fine-tune the ASPC-DA model,the recognition rate of which is 92.69％.The recognition rate of the ASPC-DA model is far better than the existing methods.Finally,the results of sensitivity experiment show that the best training sample size of ASPC-DA model is 70％of the total sample size,the best optimal iteration upper limit is from 90 to 100 and,the best optimal termination condition threshold is from 10 5 to 10-4.
关键词
交通运输规划与管理/假冒绿通车识别/机器学习/自步学习/数据增强Key words
transportation planning and management/fake toll-free vehicles/machine learning/self-pace learning/data augmen-tation引用本文复制引用
基金项目
四川省省级科研院所基本科研业务费支持项目(2021JBKY09)
四川省省级科研院所基本科研业务费支持项目(2021JBKY08)
出版年
2024
NETL
NSTL
万方数据