Accurate 3D instance segmentation is foundational for perception and decision making in embodied systems, yet prevailing approaches depend on densely annotated 3D point clouds, which are costly to acquire and difficult to scale across sensors and environments. We address this bottleneck with Fusion3DGS, an end-to-end, label efficient framework that couples 3D Gaussian Splatting with coordinated 2D–3D neural processing. From multi-view RGB images equipped only with 2D instance masks, our method optimizes a compact anisotropic Gaussian scene representation and performs instance reasoning via an occlusion aware cross-attention fusion stack. The weight-sharing lock imposes shape-consistent, gated coupling of early 2D and 3D kernels with conservative feedback and drift regularization to stabilize 2D-mask training and improve label efficiency, and a rendering consistency objective ties the Gaussian geometry to 2D supervision, enhancing boundary fidelity under occlusion and view changes. The ability to learn from widely available RGB data without dense 3D labels makes the approach practical for large-scale deployment.

In the era when drone technology is highly advanced, it is necessary to address the safety issues arising from UAV crashes under current conditions. This paper first employs a fuzzy Bayesian network to identify the key factors contributing to UAV crashes. Subsequently, relevant experts in the civil aviation field are invited to provide fuzzy linguistic evaluations of each risk factor, from which the prior probabilities of root nodes are obtained. The prior probabilities of non-root nodes are derived using a judgment matrix. Then, a Bayesian network model for UAV crash risk is constructed using GeNIe software, followed by single-factor analysis, as well as forward and backward reasoning to determine the critical factors influencing UAV crashes. Finally, an Event Sequence Diagram (ESD) is used to analyze the consequences of UAV crashes, providing provides ideas for UAV crash risk assessment and post-accident consequence control methods.

Drones have, in recent years, been considered as a promising way for the last-mile delivery of urban logistics. In view of limited or short battery life of drones in every operation, the drone recharging stations and their locations are crucial to ensure delivery accessibility and to accomplish the required mission. Therefore, in addition to considering the selection of service centres and demand nodes, the location of recharging stations should also be incorporated into the proposed model of hierarchical facility location problem (HFLP). Many existing researches on drone recharging station locations assume that the delivery demand is known and only a single path is designed to meet the demand by using queuing theory models to describe node-based service waiting time. However, the reality is that we are still in the infrastructure planning stage for drone delivery, in which the demand is unknown in most cases. Our research proposes a novel mix-integer programming model to solve the HFLP for the last-mile drone delivery in urban areas, aiming to minimise the deployment, operation, and maintenance (DOM) costs of recharging stations, service centres, and paths. We have also incorporated a demand satisfaction constraint into the selection of effective alternative paths that meet the delivery accessibility. On the other hand, a sensitivity analysis in our study reveals varying degrees of impact on DOM costs and effective alternative path quantities, whose results would be useful to practical drone logistics delivery in urban areas.

Modelling passenger and train traffic is a significant approach to evaluate the performance of urban rail transit (URT) networks. However, the heavy computation pressure caused by high-efficiency requirements, massive passengers, and high network complexity makes it more challenging to integrate passenger and train traffic simulation into a unified model. We propose an efficient multi-agent model to simultaneously simulate passenger and train traffic in the URT network. The model framework comprises several agents, including passenger batch, train, line, and network. A passenger aggregation method is proposed to release the computation pressure. The model is tested in the URT network of Chongqing, China. The experiment results show the model can handle a 1.6 million passengers, 1900 trains simulation within 86 s, without losing any passengers' specific travel spatial and temporal trajectory. Three experiments are conducted for further validation, including analysing the transportation performance under different passenger route assignments, train headways, and AFC data, respectively.

通过构建交通运输类虚拟仿真实验教学体系,着重破解在实验教学中存在的虚拟仿真实验教学资源共享不充分,实验教学与行业产业需求匹配脱节,缺乏多岗位协同和多场景模拟以及实验内容与现有教学环节融合度不高等问题.聚焦高铁列车调度指挥、库存管理与自动化仓储、城市轨道交通调度指挥等关键问题,以点带面,重点打造了系列国家级、省级高阶性虚拟仿真实验教学课程,紧密融合交通运输类专业课程体系,构建符合交通运输类的人才培养模式,实现贯通式人才培养过程,提出并实践了4个"二一"虚拟仿真实验教学方法,为虚拟仿真实验课程的教学改革与实践提供了支撑.

高速铁路调度员监控作业的注意力水平识别是全周期注意识别的重要组成部分.针对监控作业交互少、反馈弱以及视觉特征不显著的特点,设计基于信息感知密度的注意诱导实验作为客观评价参照,采集全头脑电并提取了 57个通道的7项频段指标为识别特征.采用Pearson相关系数进行特征初筛,采用Logistic回归-预测变量重要性排序的包裹式方法对特征进行进一步降维并进行注意水平识别.实验结果表明:基于左额叶和双侧枕叶的17个脑电频段特征的多项Logistic回归模型对低注意水平有81％的识别准确率.脑电频段特征对应负责大脑思维功能和视觉加工处理的脑功能区,反映高铁调度员在监控工作中的注意水平变化对应的认知功能变化.

For the classification and recognition of the working status in the warning and intervention of the fatigue and working status of high speed railway dispatchers, this paper proposed a classification and recognition method of high speed railway dispatchers’ working status based on eye-movement characteristics. Eye-movement data were collected by experimenters in the simulation experiment of the schedule job within the stage plan. The pre-set experimental tasks were used as the objective classification criteria for working status and as an influencing factor for studying the distribution law of the eye-movement characteristic index. Through discriminant analysis, a typical discriminant function was established to determine the prognosis of the dispatcher’s working status, and a discriminant threshold correction method considering misjudgment loss was proposed. Combined with the characteristics of the schedule job, a mechanism for the working status intervention time determination was established. The results show that dispatchers display signs of reduced vigilance in the monitoring tasks, and the discriminant analysis method can effectively identify the dispatcher’s working status with an average accuracy rate of 81.3%. On the basis of correcting the discrimination threshold and considering the misjudgment probability,the determination mechanism of working status intervention time effectively avoids the occurrence of wrong intervention.

高铁调度员的工作负荷与脑力认知资源的占用水平密切相关,过高的工作负荷会影响高铁调度员的工作状态,进而影响高速铁路的运营安全.因此该文设计了一种实时监控高铁调度员工作负荷的实验方法,通过识别高铁调度员的工作负荷等级,提示其及时休息或采取相应安全措施.实验以高铁调度员的主观负荷值、认知资源占用量、工作任务量3项指标为工作负荷的标定指标,以高铁调度工作内容为基础,参照oddball范式,诱发脑电事件相关点位P300成分,并从中提取时域、频域、非线性特征3类指标;以支持向量机为工作负荷识别模型,对脑电信号样本训练集数据进行训练学习,对样本训练集数据的工作负荷等级进行判定,最终得出脑电信号不同特征下的识别准确率、灵敏度数据,以此判定其识别效果.

为科学、安全、有效地管理铁路生产的各技术环节,国铁集团、集团公司与各站段出台了涉及行车组织、客运组织、货运组织和铁路技术设备的运用、管理、维修等方面的技术规章,它是铁路运营和维护安全生产的基本保障。技术规章管理系统作为技术规章管理的信息化平台,是技术规章高效管理和高效运用的关键,是铁路基础管理工作的重要组成部分。本文在研究国内外情况和其他行业规章、文件管理系统后,依据集团公司实际工作需要,对集团公司技术规章管理平台方案进行研究,提出了一个更加方便、快捷、功能齐全且整体上相互协调的技术规章管理系统功能模块方案,作为指导集团公司技术规章系统建设的理论支撑。

准确预测高速铁路调度员疲劳程度是提高调度员工作效率,保证列车运行安全的关键技术问题.针对该问题,提出一种基于K-Means聚类的高速铁路调度员疲劳程度预测方法.基于调度员人因失效概率值得出最佳疲劳分级数,在此基础上利用融合算法计算出疲劳程度分级阈值并作为输出端,面部特征信息作为输入端,建立出基于BP模式识别神经网络的高速铁路调度员疲劳程度预测模型.根据32名高速铁路调度员模拟调度任务的面部特征数据,对该模型试算.研究结果表明,疲劳程度最佳分类数为3,在考虑工作时间作为模型的输入指标时,平均误差为13.3％,最佳效果下的误差仅为9.3％.

调度员的疲劳作业是一个严重的铁路行车安全问题.如果在调度员作业过程中及时发现其过于疲劳并发出警告,可避免因疲劳造成的铁路行车安全事故.本研究基于隐马尔可夫模型(HMM)与BP(Back Propagation)神经网络,通过记录的5项面部特征指标(注视时间、平均瞳孔大小、眨眼频率、眨眼时长、哈欠频率)结合工作时间对长时间作业下的高速铁路调度员疲劳度进行判定.研究结果显示,HMM模型对时段特征数据集的判定准确率相对时刻特征数据集高,且对Ⅰ级疲劳判定表现较BP神经网络优秀,而BP神经网络对两种数据集的判定效果相似且均优于HMM模型,其对Ⅱ级,Ⅲ级疲劳有着较为准确的判定.

为研究成渝地区城际铁路网络的网络特性和脆弱性,利用复杂网络理论的拓扑规则,选取成渝地区城际铁路建设规划下的城际铁路网络进行建模,并从节点度、平均路径长度和聚类系数等指标进行网络特性分析.在此基础上,通过PageRank算法挖掘网络中的重要节点,模拟随机攻击和蓄意攻击,对网络的脆弱性和鲁棒性进行分析.结果表明:成渝城际铁路网络具有小世界和无标度网络特性;重庆、成都、自贡等为重要节点;在随机攻击下城际铁路网络呈现鲁棒性,蓄意攻击下呈现脆弱性.因此,在日常运营中应重点关注关键节点,为成渝地区城际铁路安全高效地运营提供理论参考.

As a critical foundation for train traffic management, a train stop plan is associated with several other plans in high-speed railway train operation strategies. The current approach to train stop planning in China is based primarily on passenger demand volume information and the preset high-speed railway station level. With the goal of efficiently optimising the stop plan, this study proposes a novel method that uses machine learning techniques without a predetermined hypothesis and a complex solution algorithm. Clustering techniques are applied to assess the features of the service nodes (e.g., the station level). A modified Markov decision process (MDP) is conducted to express the entire stop plan optimisation process considering several constraints (service frequency at stations and number of train stops). A restrained MDP-based stop plan model is formulated, and a numerical experiment is conducted to demonstrate the performance of the proposed approach with real-world train operation data collected from the Beijing-Shanghai high-speed railway.

铁路系统大力推行无纸质检票系统,身份证检票和二维码检票2种无纸质检票方式已经在多个枢纽站点的动车、高速铁路列车和城际列车检票口实施,新型检票闸机及其配套设施逐渐代替纸质车票检票闸机和人工检票口.基于无纸质检票的现场调研数据,计算无纸质检票排队参数并使用AnyLogic建立检票仿真模型,计算无纸质检票所需时间.对西南某客运枢纽应用该仿真模型,根据仿真结果提出在不同待检票人数时,各类检票闸机的数量和人员安排.仿真结果显示无纸质检票较纸质检票的检票时间减少13.76％～28.69％,依据该结果对现场检票时间进行分析,不断提高检票工作效率.

Signal-coordinated control systems have been widely implemented on urban arterials. By synchronizing consecutive intersections, signal coordination can significantly improve the throughput of vehicles along arterials. Considerable research has been dedicated to assessing the efficiency of coordinated signalization, while from a safety perspective there is a lack of necessary effort to identify the correlated heterogeneity between injury and property damage only (PDO) crashes. In this paper, the authors adopt a multivariate Poisson log-normal (MPLN) model to reveal the heterogeneous connectivity along coordinated arterials. In addition, spatial correlations of conditional autoregression (CAR) and multivariate conditional autoregression (MCAR) are added to the MPLN model for the purpose of calibrating the regressive results. With the surveyed arterials in Ann Arbor, Michigan, the case study shows that (1) the MPLN-MCAR model predicts the results more accurately compared to the MPLN-CAR and MPLN models, (2) the spatial correlation of injury crashes differs significantly from that of PDO crashes, and (3) there is a strongly correlated heterogeneity between injury and PDO crashes. The findings serve to provide effective countermeasures for safety planning, design, and management of signal-coordinated arterials.

以实际配送中心作业流程与管理需求为背景,依托自动分拣与智能堆垛实验室的实体设备,采用虚实结合的方式,建成了库存管理与自动化仓储虚拟仿真实验系统.分析了仿真系统的功能、设备、仿真原理及流程,完成了自动化仓储系统认知与布局、实时库存管理与全流程管理,实现了对"高集成、多环节、多品类"自动化仓储作业流程、"多岗位、多工种、多场景一体化"管理过程的虚拟仿真,为物流管理相关专业人才培养提供了强有力的支撑.

Starting from the following aspects: the activity system of college students’ extracurricular innovative competition, the guiding mechanism of competitions, the opening of laboratories, the organizational structure, the creation of students’ associations and bases for scientific and technological innovation, the competition system, the sound incentive mechanism and the creation of competition atmosphere, etc., this paper aims to cultivate the college students’ innovative ability as the core, and build a comprehensive and three-dimensional extracurricular innovative competition implementation platform with multi-level and multi-form, based on the characteristics of the discipline.Through this platform, the long-term mechanism of extracurricular innovative competition activities is formed, the sustainable development of competition activities is guaranteed, the participation and enthusiasm of students in extracurricular innovative competition activities are enhanced, the innovative ability, team cooperation consciousness and scientific literacy of students are effectively cultivated, the quality of personnel training is improved, and the reform of experimental teaching is promoted.

通过对调度员作业行为可靠度的研究,可以在一定程度上保障城市轨道交通运行的安全.基于城市轨道交通调度员工作任务,对调度员作业行为特征及行车系统特征的影响因素进行分析,通过层次分析法对影响因素进行筛选,提取出权重较高的6项指标.运用筛选得到的指标建立调度员行为可靠度BP神经网络,利用相关数据对网络模型进行训练,对不同时段的行为可靠度做出评价.结合某市城市轨道交通调度实例,验证分析发现断面能力、最小行车间隔及调度员作业行为能力对调度员作业行为可靠度影响最大,在早、晚高峰及午间调度员作业行为可靠度较低,需要重点监督.

The In accordance with the standard calculating method, the influence factors analysis of high-speed railway train headway is carried, and the technical measures of cutting down the train headway is proposed. Based on CRH380AL EMU with the speed of 300 km/h, the contrast of the train tracing interval of Beijing-Shanghai high-speed rail before and after the control is obvious, which is calculated by the traction calculation simulation software. The result proved the probability of setting the train headway as 4 min on Beijing-Shanghai high-speed rail. In addition, the experiment examine the technical measures of cutting down the train headway is scientific and reasonable.

The logistics facility location is always involved with great deals of investment. Its construction and operation also bring out a huge amount of the greenhouse gas (GHG) emission due to the consumption of building materials, energy, the running of trucks, and other logistics equipment. Particularly, trucking activities in the urban logistics networks (ULN) are a major source of GHG. This paper aims to formulate an eco-facility location model to minimize both the total cost of ULN construction and operation and the GHG emissions of truck trips. Based on the mathematical relations of GHG emissions rates and several macroscopic factors, which we obtained by multivariate regression analysis on a large set of empirical trucking data in our previous research, the data-driven emissions rates estimation function is acquired. Then, we link the estimation function of each trip purpose by various kinds of logistics facilities through a qualitative analysis. The eco-facility location problem is modeled by integrating the pure facility location model and the GHG emissions function. The problem is first converted to a biobjective mixed-integer program, and the Particle Swarm Optimization algorithm is applied to solve the model. Through experiments with real case, the effectiveness of the models and algorithms is verified. The eco-facility location model for ULN tends to obtain the environment-friendly location decision. Our analytical results also verify the hypothesis that locations of facility do impact the relevant truck-related GHG emissions, especially to transfer transport, as well as inbound and outbound freight.