青藏铁路冻土地温自动检测系统的设计与数据分析

针对影响青藏铁路安全行车的冻土温度特性及其测试问题,对比国内外的冻土研究方法,设计含上位机系统和下位机系统青藏铁路冻土地温自动检测系统。选择高精度热敏电阻为温度传感器,设计与布置测试点的深度和间隔,通过下位机定时的地温数据测试、保存与发送,实现地温数据的采集,并采用上位机的接收数据、处理与分析功能,完成地温参数变化的描述与分析,可为冻土面的铁路工程设计提供数据依据。同时,以青藏铁路某两断面地温检测为示例,将系统应用于地温断面的数据采集与分析中,描述冻土的状态对该段青藏铁路的影响。自动检测系统将为青藏铁路的安全与长期稳定运营提供重要的冻土变化信息。     

Airport privatization in congested hub–spoke networks

This paper investigates the airport privatization issue. One congested hub and two linked local airports serve symmetric hub carriers. Passengers valuate the congestion delay cost and benefit from greater frequencies. The government considers privatizing either the hub or local airports. We find that in each privatizing scenario, welfare-maximizing public airport(s) set a charge below their operating costs in order to fully coordinate the high charge of privatized airport(s). If this fiscal deficit is not allowed, each scenario causes distortion. Interestingly, the distortion—and hence welfare losses—in privatizing a hub are smaller (larger) than those in privatizing both local airports when both passengers’ valuations are small (large); this is exactly the case when privatized local airports are strategic substitutes (complements). We also surprisingly find that retaining the hub airport as public and privatizing one or both local airports achieves the same market outcomes. We further find that if all airports are privatized, welfare becomes worse than the other scenarios; the hub airport charges lower (higher) prices than local airports when both local airports are strategic substitutes (complements).     

通信网络运维保障转变研究

通过多个方面对通信企业通信网络运维保障发展转变进行研究,对通信网络运维保障转变进行分析和总结,并从中寻找未来通信网络运维保障的发展趋势。     

基于BP神经网络的隧道围岩分类

应用BP神经网络建立了隧道围岩分类模型,对其影响因素进行分析,并对模型进行优化,将训练好的分类模型运用到贵安隧道,为隧道围岩分类提供了一种简便可行的方法。     

GSM-R无线网络切换的关键指标研究

根据GSM-R无线网络切换的特点,对切换的几个关键技术指标进行重点分析,提出衡量GSM-R网络切换质量的几个关键指标,为网络优化提供关键参数。     

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。     

广州地铁网络化运营后的乘客信息发布水平提升措施

地铁步入网络化运营阶段后,客流量剧增,通达性增强,服务范围和运营信息量增大,对客运服务信息发布提出更高的要求。通过分析客运服务信息发布的现状及网络化运营后面临的风险和难点,提出优化应急信息发布渠道和范围、采用智能化手段提升信息发布水平、加强乘客多路径出行引导、新增快慢车宣传指引等提升措施,以满足乘客的知情权及出行需求,提高网络化运营服务质量。     

网联车辆并线预测与巡航控制的研究

为检测旁车道车辆驾驶员的并线意图,提升网联车辆巡航跟车的主动安全性,提出了一种基于NAR神经网络学习的迭代循环预测算法。NAR神经网络的训练样本由实际交通环境中的车辆并线数据获得,通过训练的网络预测未来一段时间内旁车的横向行驶轨迹,并根据划定的监控区域计算旁车的切入概率。同时,提出了一种考虑并线概率的跟车距离策略,并应用到网联车辆CACC系统中。结果表明,所提出的并线预测算法能精确计算出旁车的横向换道轨迹,所提出的跟车策略可提升车辆的跟车安全性。     

远程技术支援系统构建及其关键问题研究

通过综合通信网络,实现系统、全面、高效的远程知识共享和管理的装备维修远程技术支援系统,是实现一体化保障组织体系,加强一体化协同保障力量的重要组成部分.在不远的未来,随着通信技术和信号管理技术的发展,将构建安全、可靠、开放的远程技术支援通信网络平台一面向服务的软件架构(SOA,Service Oriented Achitecture).在此过程中,需研究更高效、通用的抽象知识表示,构建更科学的服务评价体系和全方位的通信网络安全体系.     

Full vehicle simulation using thermomechanically coupled hybrid neural network shock absorber model

In the case of full vehicle models, the technique of multi-body simulation (MBS) is frequently used to study their highly non-linear dynamic behaviour. Many non-linearities in vehicle models are induced by force elements like springs, shock absorbers, bushings and tires. Commonly, spline functions are used to represent the force responses of these components. If the non-linear relationships are more complicated, the spline approximations are no more accurate. An alternative approach is based on empirical neural networks which are based on the mathematical approximation of measured data. It is well known that neural networks are able to represent and predict complex component responses accurately. The aim of this paper is to perform a dynamic full vehicle simulation using a thermomechanically coupled hybrid neural network shock absorber model. In this shock absorber model, the spline approach is combined with a temperature-dependent neural network. Based on a displacement-controlled excitation on a four post test rig in the ADAMS/Car MBS software, a rugged test track is simulated. In this way, the front and rear shock absorbers are dynamically loaded with comfort-relevant frequencies in the range of 0.75-30 Hz and velocity amplitudes up to 2 m/s. By the simulation, stability of the hybrid neural network model is demonstrated. Furthermore, the damping force, the vertical acceleration of the chassis and the required simulation times are compared. The standard spline approach is used as a reference.