用于结构损伤位置识别的两种神经网络性能比较

为了克服用BP网络进行结构损伤位置识别时网络结构确定难、网络训练易陷入局部极小、训练时间长以及处理带噪声数据需要大量的误码练样本等问题，提出用SOFM网络进行结构损伤位置识别的方法。分别用SOFM网络和BP网络对一桁架结构进行损伤位置识别，通过比较两种网络的性能发现SOFM网络不但网络结构容易确定，网络训练不存在陷入局部极小的问题，BP网络只有在大量训练样本条件下才能保证网络具有较好的抗噪声能力，若训练样本不足，则BP网络的抗噪声性能较差，而SOFM网络在较少训练样本情况下即可具有良好的抗噪声性能，因而SOFM网络更适合训练样本有限备件下的结构损伤位置识别。     

基于神经网络的电动转向系统助力特性研究

基于综合目标函数的误差反向传播学习算法，设计了用于该离散化转向助力特性的BP神经网络结构，进行了离线训练，并在电动转向综合试验台上实现了在线控制。实现了全车速范围的非线性转向助力，克服了转向助力盲区。     

Application of Elementary Neural Networks and Preview Sensors for Representing Driver Steering Control Behaviour

This paper demonstrates the use of elementary neural networks for modelling and representing driver steering behaviour in path regulation control tasks. Areas of application include uses by vehicle simulation experts who need to model and represent specific instances of driver steering control behaviour, potential on-board vehicle technologies aimed at representing and tracking driver steering control behaviour over time, and use by human factors specialists interested in representing or classifying specific families of driver steering behaviour. Example applications are shown for data obtained from a driver/vehicle numerical simulation, a basic driving simulator, and an experimental on-road test vehicle equipped with a camera and sensor processing system.     

基于新奇检测技术的斜拉索状态评估

探讨了应用基于BP神经网络的新奇检测技术进行斜拉索状态评估的方法.通过对监测系统采集数据分析处理,生成训练神经网络需要的样本数据,按要求训练网络,建立了基于新奇检测技术的多阶段状态评估的神经网络模型,实现了斜拉索状态评估的两个阶段:状态预警、状态异常位置识别.状态异常位置识别采用逐步分区识别的方法,最终将损伤拉索的位置...     

电动助力转向系统控制的台架试验研究

提出了将基于模糊神经网络的PID控制策略用于电动助力转向系统中助力电机的控制。设计了电动助力转向试验台，并进行了电动助力转向系统的台架试验。试验结果证明，采用模糊神经网络控制器确定目标电流，并使用PID反馈控制器跟踪目标电流的控制策略是十分有效的，能显著提高汽车的转向轻便性和灵敏性。     

基于BP神经网络的电动助力转向器在线故障诊断研究

应用人工神经网络和推理法则,提出了一套针对电动助力转向器(EPS)、转矩转角传感器(TPS)和电子控制单元(ECU)的在线故障诊断策略。以传感器采样值作为神经网络的输入,TPS及ECU故障代码作为输出,对EPS关键部件进行在线故障诊断。通过EPS试验台,获取相应的数据样本,利用Matlab神经网络工具箱,对该策略进行了仿真分析,结果验证了故障诊断策略的有效性。     

基于MM-STConv的端到端自动驾驶行为决策模型

针对现有端到端自动驾驶模型输入数据类型单一导致预测精确度低的问题，选取RGB图像、深度图像和车辆历史连续运动状态序列作为多模态输入，并利用语义信息构建一种基于时空卷积的多模态多任务（Multimodal Multitask of Spatial-temporal Convolution，MM-STConv）端到端自动驾驶行为决策模型，得到速度和转向多任务预测参量。首先，通过不同复杂度的卷积神经网络提取场景空间位置特征，构建空间特征提取子网络，准确解析场景目标空间特征及语义信息；其次，通过长短期记忆网络（LSTM）编码-解码结构捕捉场景时间上、下文特征，构建时间特征提取子网络，理解并记忆场景时间序列信息；最后，采用硬参数共享方式构建多任务预测子网络，输出速度和转向角的预测值，实现对车辆的行为预测。基于AirSim自动驾驶仿真平台采集虚拟场景数据，以98 200帧虚拟图像及对应的车辆速度和转向角标签作为训练集，历经10 000次训练周期、6 h训练时长后，利用真实驾驶场景数据集BDD100K进行模型的测试与验证工作。研究结果表明：MM-STConv模型的训练误差为0.130 5，预测精确度达到83.6%，在多种真实驾驶场景中预测效果较好；与现有其他主流模型相比，该模型综合场景空间信息与时间序列信息，在预测车辆速度和转向角方面具有明显的优势，可提升模型的预测精度、稳定性和泛化能力。     

基于BP神经网络的电动车电子差速器设计

针对双电机独立轮式驱动电动汽车,介绍了一种汽车行驶数据测量系统;考虑汽车转向行驶时内、外侧车轮转速与转向角和车体速度之间的非线性关系,提出了一种新型结构的基于BP神经网络原理的电子差速方案;设计了训练和在线计算算法,并编写了C语言计算机程序;仿真验证了该差速器的可行性。     

汽车半主动悬架的神经网络自适应控制

本文提出了用两个线性神经网络对汽车半主动悬架系统进行在线辨识和控制的策略，介绍了该控制系统中神经网络的在线训练方法，进行了仿真计算和结果分析。     

BP和RBF神经网络在边坡稳定性评估中的比较研究

通过实例分析,对BP神经网络和RBF神经网络在边坡稳定性评估中的应用进行了比较研究,结果表明,BP神经网络和RBF神经网络均能很好地对边坡稳定性进行评估,但RBF神经网络比BP神经网络的训练速度更快,效率更高,并且对于同样的精度要求,RBF神经网络对边坡稳定性的评估结果更加准确和适用。     

交通量预测的神经网络集成方法

首次将神经网络集成技术引入交通量预测。神经网络集成通过训练多个神经网络并将各网络输出进行合成,可显著提高学习系统的泛化能力。在Boosting和Bagging集成方法的基础上,提出基于分治策略的神经网络集成方法,并且讨论了网络权重分配算法。使用上述三种神经网络集成预测模型,对苏州某交叉口实时交通量进行预测,预测结果比较理想,优于单一神经网络预测方法。实验表明,神经网络集成用于交通量预测是有效可行的。     

基于遗传-径向基函数神经网络的公路隧道围岩定级方法

围岩分级准确与否直接关系到隧道的施工安全和工程造价。针对现阶段围岩分级方法存在的主要问题,结合宁绩高速公路隧道群施工期围岩定级实践,以国标BQ分级为基准,在大量现场测试和室内试验的基础上,引入径向基函数神经网络,并以分级结果作为遗传-径向基函数神经网络的训练样本,建立了隧道围岩分级的遗传-径向基函数神经网络模型。应用实例表明,该模型分级结果与现场勘测基本一致,为隧道围岩分级提供了一种新方法。     

基于变权重系数的分布式驱动无人车轨迹跟踪

为了提高分布式无人车轨迹跟踪的精度，提出了基于自主与差动协调转向控制的轨迹跟踪方法。首先，在车辆三自由度模型基础上，基于模型预测控制（MPC）实时计算前轮转角以控制车辆进行自主转向轨迹跟踪。在此过程中，为了提高自主转向下车辆的轨迹跟踪精度与行驶的稳定性，考虑多种因素，利用经验公式及神经网络控制对MPC的预瞄步数和预瞄步长进行多参数调整，实现预瞄时间的自适应控制。其次，在恒转矩需求的情况下，以轨迹偏差为PID控制器的输入及左右轮毂电机转矩为输出进行差动转向控制，实现了差动转向下的轨迹跟踪控制。然后，通过设置权重系数的方法将自主与差动转向相结合。考虑到车辆横纵向动力学因素，采用模糊控制及经验公式对权重系数进行了调整，从而在提高车辆转向灵活性与轨迹跟踪效果的同时保证车辆行驶的稳定性。CarSim与Simulink联合仿真以及实车试验结果表明：与自主转向轨迹跟踪相比，采用变权重系数的协调控制可以在不同的工况下提高车辆的转向灵活性与轨迹跟踪的精度，轨迹跟踪偏差的均方根值改善率达到了11%。所提出的协调转向控制方法可为分布式驱动车辆转向灵活性的提高及轨迹跟踪精度的改善提供一种新的思路。     

Detection of Cognitive and Visual Distraction Using Radial Basis Probabilistic Neural Networks

This paper suggests a real-time method for detecting a driver’s cognitive and visual distraction using lateral driving performance measures. The algorithm adopts radial basis probabilistic neural networks (RBPNNs) to construct classification models. In this study, combinations of two driving performance data measures, including the standard deviation of lane position (SDLP) and steering wheel reversal rate (SRR), were considered as measures of distraction. Data for training and testing the RBPNN models were collected under simulated conditions in which fifteen participants drove on a highway. While driving, they were asked to complete auditory recall tasks or arrow search tasks to create cognitively or visually distracted driving periods. As a result, the best performing model could detect distraction with an average accuracy of 78.0 %, which is a relatively high accuracy in the human factors domain. The results demonstrated that the RBPNN model using SDLP and SRR could be an effective distraction detector with easy-to-obtain and inexpensive inputs.     

水上交通流冲突严重度BP神经网络评价方法

目前的水上交通流评价方法在评价指标关系模糊、来源不清等情况下难以运用,且主观性较强,存在评价结果严重偏离实际的情况,忽视了客观性不足的问题.为降低专家主观性对水上交通流冲突严重度评价的影响,基于BP神经网络建立评价模型,并通过网络训练进行函数比较,确定最符合模型设定要求的Trainlm函数,以及精度与迭代次数.由于数据的差异性会对BP神经网络的训练效率和评价精度造成影响,基于聚类分析与BP神经网络建立新的评价模型,将训练数据按照欧几里得度量进行归类开展神经网络训练,分别对水上交通流冲突严重度进行评价.运用9个水道数据为例对模型进行验证,通过比较聚类分析数据与未处理的原始数据在BP神经网络中的评价结果,发现评价结果平均误差从42.05%降低到23.74%,进一步验证了BP神经网络在该领域的可行性.评价模型利用聚类分析与BP神经网络相结合的方法,不仅客观性较强,而且与单一使用BP神经网络的模型相比提升了评价精度.      

Handling Capabilities of Vehicles in Emergencies Using Coordinated AFS and ARMC Systems

In this paper, an advanced control technique that can be implemented in hard emergency situations of vehicles is introduced. This technique suggests integration between Active Front Steering (AFS) and Active Roll Moment Control (ARMC) systems in order to enhance the vehicle controllability. For this purpose, the AFS system applies a robust sliding mode controller (SMC) that is designed to influence the steering input of the driver by adding a correction steering angle for maintaining the vehicle yaw rate under control all the time. The AFS system is then called active-correction steering control. The ARMC system is designed to differentiate the front and rear axles' vertical suspension forces in order to alter the vehicle yaw rate and to eliminate the vehicle roll motion as well. Moreover, the operation of the SMC is based on tracking the behavior of a nonlinear 2-wheel model of 2-DOF used as a reference model. The 2-wheel model incorporates real tire characteristics, which can be inferred by the use of trained neural networks. The results clearly demonstrate the enhanced characteristics of the proposed control technique. The SMC with the assistance of the ARMC provides less correction of the steering angle and accordingly reduces the possibility of occurrence of the saturation phenomenon that is likely to take place in the operation of the SMC systems.     

Handling Capabilities of Vehicles in Emergencies Using Coordinated AFS and ARMC Systems

In this paper, an advanced control technique that can be implemented in hard emergency situations of vehicles is introduced. This technique suggests integration between Active Front Steering (AFS) and Active Roll Moment Control (ARMC) systems in order to enhance the vehicle controllability. For this purpose, the AFS system applies a robust sliding mode controller (SMC) that is designed to influence the steering input of the driver by adding a correction steering angle for maintaining the vehicle yaw rate under control all the time. The AFS system is then called active-correction steering control. The ARMC system is designed to differentiate the front and rear axles' vertical suspension forces in order to alter the vehicle yaw rate and to eliminate the vehicle roll motion as well. Moreover, the operation of the SMC is based on tracking the behavior of a nonlinear 2-wheel model of 2-DOF used as a reference model. The 2-wheel model incorporates real tire characteristics, which can be inferred by the use of trained neural networks. The results clearly demonstrate the enhanced characteristics of the proposed control technique. The SMC with the assistance of the ARMC provides less correction of the steering angle and accordingly reduces the possibility of occurrence of the saturation phenomenon that is likely to take place in the operation of the SMC systems.     

基于时序-神经网络的车辆变速器齿轮故障诊断

采用时序分析和BP神经网络，建立了基于时序-神经网络的车辆变速器齿轮故障诊断系统。通过对车辆变速器齿轮运行状念特征信号进行时序分析和特征向量提取，并以此作为BP神经网络的输入向量进行网络训练，从而实现变速器齿轮运行状态的识别与故障诊断。该系统应用于LC5T81变速器齿轮的故障诊断中，能够比较准确地识别与诊断出变速器齿轮的跑合运行状态、磨损运行状态和故障运行状态。验证表明该诊断系统有效、可行。     

Adaptive Control of 4WS System by Using Neural Network

An adaptive control system of the model following type is proposed for drive motion control of a four wheel steering (4WS) car with using neural network (NN) which has mastered nonlinear friction force between tire and road surface. A model of one rigid body is adopted which represents appropriately two kinds of car motion caused by steering action, namely the lateral displacement and the yawing rotation, and an equation of motion is described in a simplified form to make a system equation for motion control possible. Nonlinear relation between the cornering force of tire and the slip angle is obtained by numerical analysis with the tire model proposed by E. Fiala, taking friction coefficient and car speed as the parameters. The result is used as the teaching signal for NN. Three NN are used in the control system composed of both the feed-forward and the feedback circuits in order to realize adaptive control. Validity and usefulness of the proposed adaptive control system with NN are verified by three kinds of computer simulation.