轨迹数据驱动的单点信号交叉口运行效率评价方法

为解决信号交叉口运行效率评价方法偏理论化、实用性不强等问题,以出租车、公交车、驾图(车联网)多源GPS轨迹数据为基础,充分利用车辆减速、停车、加速等连续速度变化特征及位置信息,提出交叉口个体车辆排队长度、通行时间、停车次数等交通参数提取技术.基于此,构建以信号交叉口运行指数为一级指标,车辆平均通行时间,第95％分位排队...     

基于LSTM网络的驾驶意图识别及车辆轨迹预测

自动驾驶汽车需具备预测周围车辆轨迹的能力,以便做出合理的决策规划,提高行驶安全性和乘坐舒适性。运用深度学习方法,设计了一种基于长短时记忆（LSTM）网络的驾驶意图识别及车辆轨迹预测模型,该模型由意图识别模块和轨迹输出模块组成。意图识别模块负责识别驾驶意图,其利用Softmax函数计算出驾驶意图分别为向左换道、直线行驶、向右换道的概率;轨迹输出模块由编码器-解码器结构和混合密度网络（MDN）层组成,其中的编码器将历史轨迹信息编码为上下文向量,解码器结合上下文向量和已识别的驾驶意图信息预测未来轨迹;引入MDN层的目的是利用概率分布来表示车辆未来位置,而非仅仅预测一条确定的轨迹,以提高预测结果的可靠性和模型的鲁棒性。此外,将被预测车辆及其周围车辆组成的整体视为研究对象,使模型能够理解车-车间的交互式行为,响应交通环境的变化,动态地预测车辆位置。使用基于真实路况信息的NGSIM（Next Generation SIMulation）数据集对模型进行训练、验证与测试。研究结果表明：与传统的基于模型的方法相比,基于LSTM网络的轨迹预测方法在预测长时域轨迹上具有明显的优势,考虑交互式信息的意图识别模块具备更高的预判性和准确率,且基于意图识别的轨迹预测能降低预测轨迹与真实轨迹间的均方根误差,显著提高轨迹预测精度。     

不同发射参数下潜射鱼雷初始弹道建模与仿真

为研究潜射鱼雷发射参数设定对鱼雷初始弹道的影响,建立鱼雷初始弹道数学模型,对鱼雷在纵平面的初始弹道进行了研究,利用Simulink编程进行了仿真,得出在不同发射参数设定下弹道的袋深、最大超调量和到达设定深度所需时间.仿真结果表明,鱼雷横舵管制舵角和出管速度对鱼雷初始弹道都有影响,其中鱼雷横舵管制舵角影响较大,为进一步优...     

桩网复合地基桩顶土拱形态分析

桩网复合地基垫层与填土中的土拱对桩顶应力调节和沉降控制效果起着重要作用.在对比现有桩网复 合地基桩顶土拱效应计算方法的基础上,通过单桩数值模拟和数学解析,对桩顶土拱形态进行分析.结果表明,桩网复合地基桩顶土拱区域内的主应力迹线是一族自桩顶平面向上曲率逐渐增大的变曲率曲线,其中,接近桩帽边缘的一条主应力迹线线形近似于抛物...     

基于移动导航数据的信号配时反推

在交通管理和评价时,信号配时对监测评价路口运行状态,评价路口配时方案至关重要.但是,大范围的实时信号配时方案的获取尚缺乏简明有效的途径.本文提出两种基于移动导航数据计算固定配时路口信号配时的方法.第一种方法是在不考虑驾驶员驾驶行为差异性时,得到路口红灯和车均延误的关系模型,从而计算某相位的红灯时长.另外一种方法是基于车辆通过停止线的时间,结合本文提出的上升梯度法,得到某阶段红灯时长.本文通过实际的路口案例计算,将预测结果和已知路口的信号配时比较,表明此方法计算得到的红绿灯时长准确度较高,为后续进行路口运行状态和通行能力研究提供了数据支持.     

鱼雷内弹道特征参数估计的回归分析

论述了以建立回归方程为中心,利用有限的实验数据建立舰艇鱼雷发射内弹道特征参数（鱼雷出管速度、膛压极值）的线性回归方程。通过实验数据计算,证明这是一种舰艇鱼雷发射内弹道特征参数估算的有效方法。在回归分析法基础上,提出了如何进行实验的优化。     

Towards comfort-optimal trajectory planning and control

ABSTRACT

This paper describes a method to analyse and evaluate different trajectory planning methods and controller types for usage in automated vehicles. Its application is shown by using a novel trajectory planning approach considering comfort aspects (based on Rapidly Exploring Random Tree (RRT)), two different controllers to follow the planned path (cascade controller and flatness based controller) and a simulation method to obtain resulting lateral vehicle accelerations. The method is used to plan and drive a trajectory through a roundabout. It can be seen that the lateral accelerations of the controller-driven vehicle are in the range of the values used for planning. However, the results of both controllers show differences in lateral deviation and in smoothness of lateral accelerations. The simulation results are then compared to real-world test drives in the same roundabout. The measured lateral accelerations are in the same range as well but show a smoother progression than the two controller models.     

Design of rollover prevention controller with linear matrix inequality-based trajectory sensitivity minimisation

This paper presents a method to design a rollover prevention controller for vehicle systems. The vehicle rollover can be prevented by a controller that minimises the lateral acceleration and the roll angle. Rollover prevention capability can be enhanced if the controlled vehicle system is robust to the variation of the height of the centre of gravity and the speed of the vehicle. For this purpose, a robust controller is designed with linear matrix inequality-based trajectory sensitivity minimisation. Differential braking and active suspension are adopted as actuators that generate yaw and roll moments, respectively. The newly proposed method is shown to be effective in preventing rollover by the simulation on a non-linear multibody dynamic simulation software, CarSim^®.     

Application of time-variant predictive control to modelling driver steering skill

The paper addresses the need for improved mathematical models of human steering control. A multiple-model structure for a driver's internal model of a nonlinear vehicle is proposed. The multiple-model structure potentially offers a straightforward way to represent a range of driver expertise. The internal model is combined with a model predictive steering controller. The controller generates a steering command through the minimisation of a cost function involving vehicle path error. A study of the controller performance during an aggressive, nonlinear steering manoeuvre is provided. Analysis of the controller performance reveals a reduction in the closed-loop controller bandwidth with increasing tyre saturation and fixed controller gains. A parameter study demonstrates that increasing the multiple-model density, increasing the weights on the path error, and increasing the controller knowledge range all improved the path following accuracy of the controller.     

人机混驾环境下基于深度学习的车辆切入轨迹预测

为保证人机混驾环境下自动驾驶汽车的安全行驶,对周围车辆切入轨迹的预测至关重要.本文首先使用Savitzky-Golay滤波器对大规模采集的自然驾驶数据进行去噪处理,根据准则提取车辆切入片段,建立符合中国道路状况的车辆切入数据集.其次,发挥Bi-LSTM网络能充分利用上下文信息的优点和in-out快捷连接有效减少梯度消失...