基于NAR神经网络的车速预测及应用
发布时间：2018-07-14 12:21
【摘要】：车速预测作为车辆智能化的重要组成部分,可为车辆的决策系统提供未来的行驶数据,对智能车辆、安全辅助驾驶及动力系统控制等研究有着重要意义。由于车速受多种因素的影响,具有显著的时变性与非线性,所以对预测有较高的要求。本文以本车车速预测为研究对象,分析车速数据时间序列特性,利用NAR神经网络在处理非线性与时变性时间序列上的优势建立预测模型对车速进行预测,并将建立的预测模型应用于防碰撞预警系统,对预测方法的有效性进行验证。本文首先通过车载OBD-Ⅱ设备与单目视觉相机采集本车车速数据及前车距离数据,并通过卡尔曼滤波对采集的数据进行滤波,为神经网络的训练提供数据支持。然后,建立了基于车速自回归的NAR网络结构,并通过反向传播算法以串联的形式对网络参数训练优化,通过与HMM车速预测方法的对比以及城市公交车工况的预测分析,验证NAR网络具有很好的预测精度、时变性能和长期预测能力。最后,将NAR神经网络预测算法应用在防碰撞预警系统中,以临界跟车安全距离模型为碰撞判断依据,预测模型预测的车速及车距用于计算临界车距,从而将预警时间提前。试验结果表明NAR神经网络预测模型可对车速有效预测。
[Abstract]:Vehicle speed prediction, as an important part of vehicle intelligence, can provide future driving data for vehicle decision-making system. It is of great significance for research on intelligent vehicles, safety auxiliary driving and power system control. Because the speed is influenced by many factors, it has significant time variability and nonlinearity, so it has a higher prediction. In this paper, this paper takes the vehicle speed prediction as the research object, analyzes the characteristics of the speed data time series, and uses the NAR neural network to predict the speed of the vehicle with the advantages of the nonlinear and time-varying time series, and applies the prediction model to the anti-collision warning system, and verifies the effectiveness of the prediction method. In this paper, the vehicle speed data and the front car distance data are collected through the vehicle OBD- II equipment and the monocular vision camera, and the data are filtered through Calman filter to provide data support for the training of the neural network. Then, the NAR network structure based on auto regression is established, and the back propagation algorithm is used in series. In the form of network parameter training optimization, through comparison with the HMM speed prediction method and the prediction and analysis of urban bus conditions, it is proved that the NAR network has good prediction accuracy, time-varying performance and long-term prediction ability. Finally, the NAR neural network prediction algorithm is applied to the collision avoidance warning system, and the critical distance model of the critical car heel is used. For the basis of collision judgment, the speed and distance predicted by the model are used to calculate the critical distance, so the early warning time is advanced. The experimental results show that the NAR neural network prediction model can effectively predict the speed of the vehicle.
【学位授予单位】：大连理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：U463.6;TP183
本文编号：2121626
[Abstract]:Vehicle speed prediction, as an important part of vehicle intelligence, can provide future driving data for vehicle decision-making system. It is of great significance for research on intelligent vehicles, safety auxiliary driving and power system control. Because the speed is influenced by many factors, it has significant time variability and nonlinearity, so it has a higher prediction. In this paper, this paper takes the vehicle speed prediction as the research object, analyzes the characteristics of the speed data time series, and uses the NAR neural network to predict the speed of the vehicle with the advantages of the nonlinear and time-varying time series, and applies the prediction model to the anti-collision warning system, and verifies the effectiveness of the prediction method. In this paper, the vehicle speed data and the front car distance data are collected through the vehicle OBD- II equipment and the monocular vision camera, and the data are filtered through Calman filter to provide data support for the training of the neural network. Then, the NAR network structure based on auto regression is established, and the back propagation algorithm is used in series. In the form of network parameter training optimization, through comparison with the HMM speed prediction method and the prediction and analysis of urban bus conditions, it is proved that the NAR network has good prediction accuracy, time-varying performance and long-term prediction ability. Finally, the NAR neural network prediction algorithm is applied to the collision avoidance warning system, and the critical distance model of the critical car heel is used. For the basis of collision judgment, the speed and distance predicted by the model are used to calculate the critical distance, so the early warning time is advanced. The experimental results show that the NAR neural network prediction model can effectively predict the speed of the vehicle.
【学位授予单位】：大连理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：U463.6;TP183
【引证文献】
相关期刊论文 前1条
1 魏久哲;王小勇;黄长宁;庄绪霞;;应用NAR运动估计的序列帧间匹配技术[J];航天返回与遥感;2017年03期
相关硕士学位论文 前1条
1 路广明;基于出行里程预测的插电式混合动力汽车控制策略研究[D];吉林大学;2017年
，本文编号：2121626
本文链接：https://www.wllwen.com/kejilunwen/zidonghuakongzhilunwen/2121626.html
