基于改进极限学习机的公交站点 短时客流预测方法

以公交车IC 卡和GPS数据为基础，提出了一种基于改进粒子群算法优化极限学习机(IPSO-ELM)的公交站点短时客流预测模型.依托IC 卡和GPS 数据在站点的特征表现和内在联系，定义了站点间距，并分析了站间距和车辆到总站距离间的联系；提出了公交乘客上车站点确定方法，进而得到公交站点上车客流量；通过分析公交客流数据特征，确定ELM输入参数维度，并采用IPSO 算法找到ELM的最优隐含层节点参数；最后依托广州市19 路公交车客流数据仓库进行了方法验证.结果表明：所用优化后的ELM方法预测误差在10%以内，并与应用广泛的SVM、ARIMA和传统ELM模型进行对比分析，发现改进的ELM方法拥有更高的可靠性和泛化性能.     

基于强化学习的干线信号混合协同优化方法

交通拥堵已成为很多大中城市普遍存在的社会问题。信号控制作为缓堵保畅的重要措施之一，愈发受到社会关注。信号优化手段可分为模型驱动和数据驱动两类，且随着交通大数据的不断充实，基于强化学习的数据驱动方法日益成为新兴发展方向。然而，现有数据驱动类研究主要偏重于决策模型设计，缺乏对智能体结构的探讨；同时，在多路口协同方面多采用分布式策略， 忽略了智能体之间信息交互，无法保障区域层面的整体最优性。为此，本文以干线信号为对象， 构建一种多智能体混合式协同决策的信号优化方法。首先，针对交通状态的多样性、异构性及数据不均衡性，设计分布训练-分区记忆的单智能体决策模型，并优化状态空间和回报函数，界定单路口控制的最佳方案；其次，融合分布式和集中式学习的模型优势设计多智能体交互方法，在单路口分布式控制的基础上，设置中心智能体评价局部智能体的决策行为并反馈附加回报以调整局部智能体的决策模型，实现干线多信号的协同运行。最后，搭建仿真平台完成效果测试与算法对比。结果表明：新方法与独立优化和分布式协同相比，在支路交通流基本不受影响的前提下， 干线停车次数分别降低了14.8%和13.6%，具有更好的控制效果。     

A modified reinforcement learning algorithm for solving coordinated signalized networks

This study proposes Reinforcement Learning (RL) based algorithm for finding optimum signal timings in Coordinated Signalized Networks (CSN) for fixed set of link flows. For this purpose, MOdified REinforcement Learning algorithm with TRANSYT-7F (MORELTRANS) model is proposed by way of combining RL algorithm and TRANSYT-7F. The modified RL differs from other RL algorithms since it takes advantage of the best solution obtained from the previous learning episode by generating a sub-environment at each learning episode as the same size of original environment. On the other hand, TRANSYT-7F traffic model is used in order to determine network performance index, namely disutility index. Numerical application is conducted on medium sized coordinated signalized road network. Results indicated that the MORELTRANS produced slightly better results than the GA in signal timing optimization in terms of objective function value while it outperformed than the HC. In order to show the capability of the proposed model for heavy demand condition, two cases in which link flows are increased by 20% and 50% with respect to the base case are considered. It is found that the MORELTRANS is able to reach good solutions for signal timing optimization even if demand became increased.     

Analyzing passenger train arrival delays with support vector regression

We propose machine learning models that capture the relation between passenger train arrival delays and various characteristics of a railway system. Such models can be used at the tactical level to evaluate effects of various changes in a railway system on train delays. We present the first application of support vector regression in the analysis of train delays and compare its performance with the artificial neural networks which have been commonly used for such problems. Statistical comparison of the two models indicates that the support vector regression outperforms the artificial neural networks. Data for this analysis are collected from Serbian Railways and include expert opinions about the influence of infrastructure along different routes on train arrival delays.     

混合式教学在《汽车测试技术》课程中的应用探索与反思

混合式教学是将线上教学与线下教学相结合的新型教学模式,充分利用混合式教学的优势以达到提升课程质量的目的。文章以《汽车测试技术》课程为例,对课程混合式教学的设计与实践进行了探索,分析了实施效果,进而总结了教学反思,旨在指导课程教学质量持续改进,同时为同类课程的混合式教学提供一定参考。     

考虑驾驶风格差异的高原公路危险路段识别研究

针对高原环境中驾驶人风格、生理变化与危险路段特征之间的潜在关联，提出一种基于驾驶状态的危险路段识别方法，辨识和分析不同风格驾驶人具有潜在风险的路段，并提出优化方案。首先，通过实车实验采集驾驶人行为及生理指标数据，使用DBSCAN(Density Based Spatial Clustering of Applications with Noise)得出驾驶风格类型，并依据行为特征对驾驶风格进行差异性 分析；其次，采用卷积神经网络、双向长短时记忆神经网络与注意力机制搭建危险状态识别模型，通过GPS(Global Positioning System)点位对应实现危险路段辨识，并基于驾驶风格差异，从驾驶人感知、操纵与生理角度对危险路段进行致因分析；最后，将生理与道路线形作为优化参考，以车速建议为着力点进行多元回归分析，并按照生理舒适域确定车速建议区间。结果表明：驾驶人根据行为特点分为谨慎、稳健和激进型，3类驾驶人在上行和下行途中的危险路段多为具有弯坡特征的组合型路段；海拔提升可加速危险驾驶状态的出现，各类驾驶人在上行时的紧张状态多源于弯坡组合值和转角值的增长，激进型驾驶人在坡度大于6%的直纵坡路段时亦会开始高度紧张；下行时，谨慎与激进型驾驶人在直纵坡坡度大于3%时易出现危险状态，激进型驾驶人在转角值大于80°且弯坡组合值大于50时亦存在驾驶风险。研究成果可满足高原公路人因事故预防的需求，为线形设计与交通管理措施制定提供理论依据。     

Towards data-driven car-following models

Car following models have been studied with many diverse approaches for decades. Nowadays, technological advances have significantly improved our traffic data collection capabilities. Conventional car following models rely on mathematical formulas and are derived from traffic flow theory; a property that often makes them more restrictive. On the other hand, data-driven approaches are more flexible and allow the incorporation of additional information to the model; however, they may not provide as much insight into traffic flow theory as the traditional models. In this research, an innovative methodological framework based on a data-driven approach is proposed for the estimation of car-following models, suitable for incorporation into microscopic traffic simulation models. An existing technique, i.e. locally weighted regression (loess), is defined through an optimization problem and is employed in a novel way. The proposed methodology is demonstrated using data collected from a sequence of instrumented vehicles in Naples, Italy. Gipps’ model, one of the most extensively used car-following models, is calibrated against the same data and used as a reference benchmark. Optimization issues are raised in both cases. The obtained results suggest that data-driven car-following models could be a promising research direction.     

Analysis of air traffic control operational impact on aircraft vertical profiles supported by machine learning

The Air Traffic Management system is under a paradigm shift led by NextGen and SESAR. The new trajectory-based Concept of Operations is supported by performance-based trajectory predictors as major enablers. Currently, the performance of ground-based trajectory predictors is affected by diverse factors such as weather, lack of integration of operational information or aircraft performance uncertainty.Trajectory predictors could be enhanced by learning from historical data. Nowadays, data from the Air Traffic Management system may be exploited to understand to what extent Air Traffic Control actions impact on the vertical profile of flight trajectories.This paper analyses the impact of diverse operational factors on the vertical profile of flight trajectories. Firstly, Multilevel Linear Models are adopted to conduct a prior identification of these factors. Then, the information is exploited by trajectory predictors, where two types are used: point-mass trajectory predictors enhanced by learning the thrust law depending on those factors; and trajectory predictors based on Artificial Neural Networks.Air Traffic Control vertical operational procedures do not constitute a main factor impacting on the vertical profile of flight trajectories, once the top of descent is established. Additionally, airspace flows and the flight level at the trajectory top of descent are relevant features to be considered when learning from historical data, enhancing the overall performance of the trajectory predictors for the descent phase.     

数据驱动跟驰模型综述

车辆跟驰模型是被交通科学与交通工程领域广泛认可的微观交通流模型，是交通流理论 的基础。近年来，信息感知与获取、大数据、人工智能等技术快速发展，推动了数据驱动跟驰模型 的快速发展。数据驱动跟驰模型，是以真实的车辆行驶数据为基础，利用数据科学与机器学习等 理论和方法，通过样本数据的训练、学习、迭代、进化，挖掘车辆跟驰行为的内在规律。本文系统 回顾了数据驱动跟驰模型在过去20余年的发展历程以及由神经网络和深度学习带动的两次研究 热潮，归纳了基于传统机器学习理论的跟驰模型、基于深度学习的跟驰模型、模型与数据混合驱 动的跟驰模型3类数据驱动跟驰模型，并分别介绍了其中的典型代表。分析数据源发现，尽管各 种高精度轨迹数据不断涌现，目前研究仍多使用美国于2006年发布的Next Generation Simulation (NGSIM)高精度车辆轨迹数据，模型的可移植性和泛化能力值得思考与研究。提出关于模型输 入、输出的3个问题：如何考虑更多驾驶行为变量，是否有必要考虑更多行为变量，现有输入、输出 是否可替换。在模型测试与验证方面，发现并讨论了目前测试不充分、对比不完整、缺少统一测 试集与测试标准等问题。最后，探讨了数据驱动跟驰模型原创性与成功的关键因素等问题。期 望通过本文的梳理，帮助研究者更好地了解数据驱动跟驰模型的过去与现状，促进相关研究的快 速发展。     

基于轻量化网络和注意力机制的智能车快速目标识别方法

为提高智能车在真实环境中的实时检测能力，改善复杂环境下检测效果不佳的问题，本文提出一种基于轻量化网络和注意力机制的智能车快速目标识别方法。首先，为了减少网络计算参数和提升目标识别算法的推理速度，提出利用GhostNet加速YOLOv4的特征提取；其次，为了提高复杂场景下对道路目标的识别精度，在GhostNet和特征金字塔部分添加结合软阈值化改进的注意力模块；最后，为了验证本文提出方法的有效性，选取Pascal VOC、KITTI公开数据集和自制城市道路数据集进行实验对比。与其他目标检测算法在精度和速度上进行比较，结果证明，本文方法在平均检测精度提升1.7%的情况下，模型参数量降低到原来的18.7%，检测速度提升了 66%，检测速度和精度均优于其他算法，可满足智能车的实时感知需求。