基于遗传算法优化的BP神经网络跟车模型研究

发布时间：2018-06-06 04:27

本文选题：跟车模型 + 神经网络　；参考：《长安大学》2015年硕士论文

【摘要】：车辆的跟驰行为是车辆行驶中的常见驾驶行为之一,特定的驾驶人由于其跟车过程中心理感知等因素的差异,其行车车距和相对速度等安全范围均不同。如果能够对驾驶人的这种跟车特性进行模拟,建立起特定的跟车模型,就可以比较相似跟车状态下的驾驶人的跟车行为是否存在异常。采取相应的措施对危险的跟车行为及时进行预警,就能够有效降低事故的发生率。本文通过实际道路跟车试验,使用视频监控系统和毫米波雷达等试验设备对车辆跟车过程中的相关跟车数据进行采集。通过分析稳定跟车过程中影响驾驶员进行加减速操作的相关数据,进而基于筛选出有效的试验数据进行预测模型研究。本文的主要研究内容和结论:(1)通过对跟车模型的相关研究回顾,提取试验数据中与驾驶员进行加减速操作相关的车辆运行参数和道路环境数据等。分析初步得出车辆相对速度、相对距离和本车速度可作为模型的特征输入参数。(2)通过对初步提取的特征参数数据进行处理,建立起BP神经网络跟车模型。通过预测分析可知BP模型很容易陷入局部极值点,而这种问题不能通过其自身结构的优化解决,因此考虑使用遗传算法来对其进行优化。(3)遗传算法对BP神经网络模型进行结构优化后,结果表明以车辆相对速度、相对距离和本车速度为组合的跟车行为预测模型准确度最高,但也仅为90.29%。通过反复试验可知经过遗传算法优化后的双隐含层的BP神经网络能够将预测准确率提高到94.17%。结果表明遗传算法优化后的双隐含层BP神经网络跟车模型,能够对车辆的跟车状态进行很好地预测。本研究得到了教育部长江学者与创新团队支持计划项目(IRT1286)和交通运输部应用基础研究项目(2013319812150)的资助。
[Abstract]:Car-following behavior is one of the common driving behaviors in vehicle driving. Because of the difference of psychological perception in the process of following the vehicle, the safety range of driving distance and relative speed are different. If we can simulate the characteristics of the driver and set up a specific model, we can compare whether the behavior of the driver in the similar state is abnormal or not. It can effectively reduce the incidence of accidents by taking corresponding measures to warn the dangerous following behavior in time. In this paper, we use video surveillance system and millimeter-wave radar to collect the data of vehicle following through the actual road following test. Through the analysis of the relevant data which affect the driver's acceleration and deceleration operation in the process of stable following the vehicle, the prediction model is studied based on the screening of effective test data. The main contents and conclusions of this paper are as follows: (1) based on the review of the related research on the following model, the vehicle operation parameters and the road environment data related to the driver's acceleration and deceleration operation are extracted from the test data. The relative speed, relative distance and vehicle speed of the vehicle can be regarded as the characteristic input parameters of the model. The BP neural network model is established by processing the data of the initial extracted characteristic parameters. The prediction analysis shows that BP model is easy to fall into local extremum, but this problem can not be solved by optimizing its own structure. Therefore, considering the use of genetic algorithm to optimize the structure of BP neural network model, the results show that the combination of vehicle relative speed, relative distance and vehicle speed is the most accurate model for prediction of car-following behavior. But only 90.29. Through repeated experiments, it can be seen that the BP neural network with double hidden layers optimized by genetic algorithm can improve the prediction accuracy to 94.17%. The results show that the double hidden layer BP neural network model, which is optimized by genetic algorithm, can well predict the vehicle following state. This study was supported by the Ministry of Education's Yangtze River Scholars and Innovation team support Project (IRT1286) and the Ministry of Transport's Applied basic Research Project (2013319812150).
【学位授予单位】：长安大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U491.255;TP18

【参考文献】