Integrated solution for anomalous driving detection based on BeiDou/GPS/IMU measurements

There has been an increasing role played by Global Navigation Satellite Systems (GNSS) in Intelligent Transportation System (ITS) applications in recent decades. In particular, centimeter/decimetre positioning accuracy is required for some safety related applications, such as lane control, collision avoidance, and intelligent speed assistance. Lane-level Anomalous driving detection underpins these safety-related ITS applications. The two major issues associated with such detection are (1) accessing high accuracy vehicle positioning and dynamic parameters; and (2) extraction of irregular driving patterns from such information. This paper introduces a new integrated framework for detecting lane-level anomalous driving, by combining Global Positioning Systems (GPS), BeiDou, and Inertial Measurement Unit (IMU) with advanced algorithms. Specifically, we use Unscented Particle Filter (UPF) to perform data fusion with different positioning sources. The detection of different types of Anomalous driving is achieved based on the application of a Fuzzy Inference System (FIS) with a newly introduced velocity-based indicator. The framework proposed in this paper yield significantly improved accuracy in terms of positioning and Anomalous driving detection compared to state-of-the-art, while offering an economically viable solution for performing these tasks.     

Parallel computing with domain decomposition for the vehicle crashworthiness simulation

The research presented in the paper deals with explicit nonlinear finite element calculation with domain decomposition for vehicle crashworthiness simulation. This is very important for vehicle design. Parallel computing is an efficient solution method to speedup and enhance the solving ability of large-scale numerical simulation. In this paper, a cost-effective domain decomposition method based on contact balance is presented, and the algorithm flowchart including contact computing is provided, and the parallel computing process and communication overhead are analyzed. Furthermore, scalability of the parallel computing method on different hardware platforms, the SGI Onyx 3800 and the ShenWei cluster, is studied. Finally, the effect of different domain decomposition strategy on vehicle crashworthiness simulation computing efficiency is presented. To end users, the research results should provide a reference for vehicle design and choosing appropriate hardware platform and computing software.     

汽车驾驶员不安全行为评价分析

为稳步增强汽车驾驶的安全性,减少安全事故的发生,驾驶人员除了需要具备足够的应急应变能力之外,还需要做好安全行为习惯养成工作,形成良好的驾驶习惯。文章以汽车驾驶员不安全行为作出研究对象,在明确不安全行为表现的基础上,深刻分析影响汽车驾驶人员不安全行为的因素,在此基础上,制定合理的应对策略,旨在引导驾驶人员形成安全行为,以不断提升驾驶员的驾驶能力。     

Understanding the limitations of different vehicle models for roll dynamics studies

An accurate and realistic vehicle model is essential for the development of effective vehicle control systems. Many different vehicle models have been developed for use in various vehicle control systems. The complexity of these models and the assumptions made in their development depend on their application. This article looks into the development and validity of vehicle models for prediction of roll behavior and their suitability for application in roll control systems. A 14 DOF vehicle model that includes dynamics of roll center and nonlinear effects due to vehicle geometry changes is developed. The limitations, validity of simplified equations, and various modeling assumptions are discussed by analyzing their effect on the model roll responses in various vehicle maneuvers. A formulation of the popular 8 DOF vehicle model that gives good correlation with the 14 DOF model is presented. The possible limitations of the 14 DOF model compared with an actual vehicle are also discussed.     

人机混驾环境下基于LSTM的无人驾驶车辆换道行为模型

道路系统中的人机混驾交通环境是指人工驾驶车辆与自动驾驶车辆混合运行的交通环境，其中换道行为建模是人机混驾环境下无人驾驶车辆行为研究的热点。基于深度学习理论，构建人机混驾环境下基于长短期记忆神经网络的无人驾驶车辆换道行为模型（Long-short-term-memory-based Autonomous Vehicles Lane Changing，LSTM-LC）。通过研究人工驾驶车辆在换道过程中与周边车辆的相互作用，对换道行为影响因素进行分析；同时，为了提升模型的迁移性，引入道路横向偏移量信息。结合LSTM神经网络的输入要求，使用美国公开交通数据集Next Generation SIMulation（NGSIM）构建换道行为样本库。针对LSTM-LC模型，以均方差MSE作为损失函数，使用RMSprop优化方法进行训练，对LSTM网络结构、历史序列长度N及训练样本量3个重要参数进行标定。最后，针对道路横向偏移量M对LSTM-LC模型性能的影响进行对比试验。研究结果表明：相比GRU-LC模型，LSTM-LC模型对换道行为的表征更准确，在模型的精度和迁移性上有着显著的提升；GRU-LC模型的均方差为4.64 m²，迁移性均方差为119.82 m²，而LSTM-LC模型的均方差为3.18 m²，迁移性均方差为79.58 m²，分别优化了31.5%和39.71%；通过引入道路横向偏移量M，可将LSTM-LC模型精度和迁移性提升约10%，且模型稳定性更强。     

卫星定位数据驱动的营运车辆驾驶人驾驶风险评估模型

为了提高营运车辆驾驶人安全管理的精细化水平，合理地评估驾驶人驾驶风险程度，有的放矢地降低高风险驾驶人的事故率，基于卫星定位数据特点及驾驶行为与驾驶风险的相关关系设计26个驾驶行为特征参数。考虑到高速和非高速行驶时相同驾驶行为对驾驶风险的影响区别较大，根据23名营运车辆驾驶人的实测数据有针对性地筛选高速和非高速路段驾驶人风险评估指标，构建营运车辆驾驶人驾驶风险评估指标体系。然后，基于熵权法、独立性权系数法和Spearman相关系数法建立集成赋权法，确定各评估指标的权重。最后，雇佣40名营运车辆驾驶人进行实车试验以验证模型的合理性。结果表明：车辆速度和加速度方面的驾驶行为特征可以用于评估驾驶人的驾驶风险且评估效果较好，驾驶风险评估得分与实际交通冲突次数呈正相关关系，所建立模型可以较为准确地评估营运车辆驾驶人驾驶风险的高低，准确率达到77.50%，该模型在不同地区使用时，准确率存在一定的差异，但在容许范围之内，方法具有较好的鲁棒性。     

制动踏板感觉评价客观方法研究

随着中国经济的飞速发展,消费者从对车辆的交通运输的简单需求提高为对车辆安全、可靠、舒适等要求其中制动性能的优劣对于汽车安全舒适运行来说尤为重要。而制动踏板感觉的好坏直接影响着车辆的制动性能因此搭建一套可行的制动踏板感觉的客观评价方法对于企业的工程开发以及提高消费者驾乘体验有着极大的助力作用。     

基于多属性决策的无人驾驶自主变道决策技术研究

在无人驾驶领域,随着车辆上布置的传感器不断丰富,无人驾驶系统可以从周围环境获得更多的信息,从而为车辆的自主判断行为决策带来可能。然而即使是有经验的驾驶员在进行变道动作的时候也需要格外的小心,因此无人驾驶系统的变道行为分析需要做到足够准确和谨慎,才能保证安全性,这也正是目前为止还没有非常完善的无人驾驶变道决策系统的原因。本文提出一种基于多属性决策的无人驾驶自主变道决策技术,帮助无人驾驶车辆在道路行驶中进行更有效、更安全的自主变道决策。     

商用车SCR尾气后处理系统的匹配控制技术

借鉴国外排放控制的成功经验,SCR尾气后处理控制技术已成为各厂家的首选方案。本文就SCR尾气后处理控制技术的工作原理、基本组成、工作过程及整车集成与匹配设计等方面做了系统的介绍和分析,期望能帮助读者对该控制技术有一个全面的认识,并对今后的产品开发工作有所裨益。