【作者】 詹盛；

【导师】 付锐；

【作者基本信息】 长安大学 ， 交通运输规划与管理， 2014， 博士

【摘要】 现代汽车安全技术的主流趋势已经由被动安全系统转为主动安全系统。车辆主动安全系统能够在交通冲突发生的早期对驾驶员进行提示或者介入车辆的操控，从而避免交通冲突的进一步恶化而引发事故。从保证安全的角度而言，当存在交通冲突风险时，车辆主动安全系统应该尽早的作出辨识。从时间序列而言，如果能够使用车辆的现有行驶状态表征参数对车辆下一步的运动状态进行预测，则可以对“即将到来”的交通冲突进行预先准备，从而进一步提高车辆主动安全系统的生效时间。车辆的行驶状态表征参数纵多，同时车辆行驶交通环境类型也不一样，因此，如何利用现有参数对车辆行驶状态和交通环境进行预测是车辆主动安全系统在算法设计时需要重点考虑的问题。针对车辆主动安全系统对于参数预测的关键技术需求，本文利用多类型传感器搭建了车辆行驶过程中的表征参数同步采集试验平台，实现了车载环境下多车辆行驶车速、交通环境参数的同步采集。利用上述试验平台对10名被试在不同道路环境下开展了真实驾驶试验，获取了大量的车辆行驶状态真实参数。考虑车辆主动安全系统在线运行的真实特点，在对国内外现有技术进行分类总结的情况下，主要完成了以下的研究内容：1、提出了基于几何分析方法的车辆换道过程中越线时间预测模型。通过使用车辆与车道线距离数据，分析了车辆换道过程中的几何特性。并结合车-路几何模型，提出了车辆换道过程中的车辆偏航角估计理论。针对直道路段和弯道路段，并考虑车辆换道方向与道路弯道方向，分别提出了车辆在直道路段和弯道路段换道过程中的越线时间预测模型。采用真实数据对预测模型的精度进行验证，结果表明，模型整体预测误差较小，且绝大部分的误差分布于零点附件。所进行的检验结果中，直道路段预测误差绝对值小于等于0.1s的比例达到了78.3%，弯道路段相应的比例达到了80.8%，且两种模型的预测误差均符合正态分布规律。2、通过建立车-路之间的几何关系模型，并采用车速与横摆角速度对道路曲率进行估计，提出并建立了ACC系统对有效目标、潜在有效目标和无效目标的辨识理论与模型。对辨识模型分别进行了单目标追踪、多目标追踪以及多目标状态切换追踪的检验，结果表明，本文所建立的模型能够有效的区分三类目标。在此基础上，利用模糊加权评价方法，采用目标车的速度、目标车跟车时距、目标车横向运动状态等参数建立了前方车辆状态切换的预测模型。采用真实试验数据对预测模型进行检验，结果表明，该模型对目标车不同状态的切换预测准确率均超过了90%。3、针对车辆运行过程中对于自车运动状态参数的预测需求，以线性二自由度车辆模型为研究对象，采用模糊Petri理论建立了车辆运行轨迹模型，将车身横向、纵向加速度、俯仰以及侧倾角速度作为输入变量建立了车辆运行状态预测模型，分别实现了对自车运行速度、横摆角速度、运行轨迹等参数的预测。针对单纯BP神经网络模型在对车辆运动状态预测过程中存在的不足，本文提出采用贝叶斯滤波器对BP神经网络模型的结果进行优化，检验结果表明，该方法将预测准确率由83.6%提高到了92.4%。本研究得到了国家自然科学基金项目（51178053和61374196）和教育部长江学者和创新团队发展计划项目（IRT1286）的资助。更多还原

【Abstract】 Vehicle active safety system was used more frequently in different vehicles then passivevehicle safety system. During the process of traffic conflict, vehicle active system can warnthe driver or control vehicle at early stage, so severity traffic conflict or accidents can beavoid. For vehicle safety, when existing traffic conflict, vehicle active safety system shoulddistinguish it as early as possible. From a time sequence view, if the next running state ofvehicle can be predicted by uding present state, then the coming soon traffic conflict can bepredicted. Based on this, the operation time of vehicle active system can improved. Runningstate characterization parameters of vehicle and traffic environment were different acoordingto time or others, so how to predict vehicle running state and traffic environment was a keytechnology while design vehicle active safety systems.Aming at the parameters predict requirements of vehicle active safety system, differentkinds of sensors were used to establish a test vehicle for data capture, vehicle speed and otherparameters can be captured in-phase. Ten drivers were called to drving this test vehicle indifferent road conditions, and large amout of vehicle running state data were obtained.Considering the real requirement of vehicle active system, the main reaserch content of thispaper were list as following:1. Time to line crossing (TLC) predict model during lane change process was bringingforward base on geometry analysis. Distance between vehicle and lane line was used toanalyzing the geometry characteristic duing lane change. By using vehicle-road geometrymodel, yaw angle predict method of vehicle during lane change was established. Aiming atstraight road and curve road, and considering lane change direction and curve direction, TLCpredict model of straight road and curve road was obtained. Real road test data was used toanalyzing the predict accuracy. Test results shows that the predict error was limited around0.Among all test results, predict error of straight road equal or less than0.1seconds achived78.3%, the similar result of curve road was80.8%. Predict error of two models meets thenormal distribution.2. By establishing vehicle-road geometry model, speed and yawrate were used to estimatethe road curvature. Based on this, distinguish model among availability target, latency availability target, and inefficacy target of ACC system was obtained. Single target test,Multi-target test, and multi-target state exchange test were carried out to test the accuracy ofthis model. Test result shows that the model can distinguish three types target accurately.Based ont this, fuzzy weighted evaluated model of target state exchange were estblished byuding target vehicle speed, target vehicle head time, target vehicle lateral moveing and otherparameters. Test results shows that the predict accuracy of different target state exchangeexceed90%.3. Aiming at the predict requrment of own vehicle, two degrees of freedom linear vehiclemodel was used. Fuzzy Petri net theory was used to estblishe vehicle running trajectorypredict model. Vehicle lateral moving, portrait moving state, pitching angle speed, and listangle speed were treated as input variables to obtaining vehicle running state predict model.Own vehicle speed, yawrate and running trajectory and other parameters were predict. Amingat the predict shortage of BP NN model, Bayesian filters was used to optimizing the predictresults, and the predict accuracy was improved from83.6%to82.4%.The research was sponsored by National Natural Science Foundation (51178053and61374196), Chang Jiang Scholars and Innovative Team Development Plan Program of theMinistry of Education (IRT1286).更多还原