To address the challenges of feature extraction and degradation state identification for railway turnout switch machine power signals over the full life cycle, this paper proposes a multi-dimensional feature-fusion-based degradation state identification method for S700K turnout switch machines. Multi-domain features are first extracted from degradation power signals in the time domain, frequency domain, and time-frequency domain. Subsequently, a Uniform Manifold Approximation and Projection (UMAP)-based feature fusion strategy is employed to construct low-dimensional feature representations that effectively characterize the evolution of the equipment’s operating state, and corresponding degradation performance indicators are established. Based on the fused features, the K-means++ clustering algorithm is applied to divide the performance degradation process of the switch machine into different stages. The clustering results are comprehensively evaluated using the silhouette coefficient, Calinski–Harabasz (CH) index, and Davies–Bouldin (DB) index, and are compared with those obtained by the fuzzy C-means algorithm and the conventional K-means algorithm. Experimental results demonstrate that the proposed method achieves superior clustering quality and stability in degradation stage partitioning, enabling refined identification of degradation states and providing reliable theoretical support and technical foundations for condition monitoring and maintenance decision-making in intelligent railway turnout operation and maintenance systems.

Object detectors based on convolutional neural networks (CNNs) have gained widespread adoption across various visual domains. However, these detectors face significant threats from adversarial attacks. Current gradient-based adversarial attack approaches have shown promising performance in misleading classification systems. Nevertheless, it is crucial to generate adversarial examples rapidly in real-time professional domains, where fixed iteration sizes are employed. Existing literature primarily focuses on adversarial example generation methods for the entire image, overlooking domain-specific requirements. In this research, we present the Sign Momentum Gradient Method (SMGM) as a novel approach to effectively and comprehensively launch white-box attacks. Subsequently, we employ this method to improve the efficacy of attacking the You Only Look Once (YOLO) v4 object detector. SMGM leverages adaptive accumulation of perturbations to achieve this goal. Our attack algorithm encompasses object invisibility attacks and object misclassification attacks, ensuring full-scale efficacy. Additionally, within the SMGM framework, we introduce the Adaptive Perturbation Suppression (APS) technique to constrain background perturbations and adjust the perturbed space's dimensions. Through extensive experiments, we validate the superior performance of our method in both object invisibility attacks and object misclassification attacks. Our proposed method achieves a remarkable recall rate (RR) of 0.79 % when fooling YOLOv4 object detection networks against object invisibility attacks on the Pascal VOC2007(test) dataset. These results serve as a source of inspiration for further advancements in adversarial attacks targeting other computer vision tasks.

Automatic driverless technology has been increasingly deploying object detectors that adopt deep learning networks to perceive running conditions, leading to security risks for object detector systems since recent studies have pointed out the existence of adversarial examples. Existing methods are mostly limited to classifier networks and have limitations in detecting adversarial examples targeting object detection networks, lacking generalization and decision foundations. However, accurate detection of adversarial examples targeting object detection networks is crucial for the field of automatic driving. In this paper, we propose an adversarial example detection method tailored for object detection networks, utilizing image processing techniques and consistency checks. The method first processes an input image and then feeds both the original input and its processed image into object detector systems in parallel. Subsequently, we propose an Intersection over Union (IOU) computation method for adversarial example detection. By calculating the IOU value between two images processed in parallel, we can determine whether an image constitutes an adversarial example. Our method is evaluated on the traffic running environment in the real world, and the detection method reaches high accuracy. Our research will inspire further efforts in detection and defense measures against adversarial examples in other scenarios.

As an efficient public transportation mode, subways are experiencing continuous growth in passenger flow. Traditional manual scheduling models are no longer sufficient, and accurate passenger flow prediction algorithms are urgently needed to optimize operational management. This paper constructs a multi-dimensional fusion subway passenger flow prediction algorithm system based on computer vision technology. First, computer vision technology is used to collect real-time passenger data within subway stations, and deep learning algorithms are used to extract and analyze data features. Second, a spatiotemporal data fusion strategy is introduced to improve the model's adaptability to complex passenger flow patterns. Experimental results show that using parallel computing technology, the algorithm runs within 80 seconds when the data size is 1 million, providing effective support for intelligent scheduling and rational resource allocation in subway operations.

With the acceleration of urbanization, the subway, as a high-capacity public transportation tool, faces huge challenges in network passenger flow diversion. Traditional static path planning is difficult to adapt to real-time changes in passenger flow. This study constructed a dynamic model of subway passenger flow based on spatiotemporal data, combined with reinforcement learning and genetic algorithm to optimize path planning strategies, and verified the effectiveness of the algorithm through a simulation experimental platform. The results showed that after adopting dynamic algorithms, the passenger flow at Station 3 of Line A decreased from 1200 people/hour to 980 people/hour, providing innovative technological solutions for efficient passenger flow diversion in the subway.

Detecting internal defects in degrading rails to ensure transport capacity and safety in railway engineering has consistently been a critical concern. However, current ultrasonic-based detection works exhibit limitations in efficiency and accuracy, especially under limited computational resources. To address this challenge, this paper proposes a method that involves a specialized B-scan image processing pipeline and an LRID (lightweight detection model for rail internal defects). Specifically, the pipeline comprises normalization, multi-channel partition filtering and data augmentation utilizing traditional transformations and generative artificial intelligent model. The pipeline aims at enhancing signal-to-noise ratio and learnability of the B-scan dataset. For constructing the LRID, a scaled version of RTDETR-L (real-time detection transformer–large) is leveraged as the baseline. The encoder of the baseline is first pruned to reduce complexity. Then, a new convolutional block called GRG (ghost-rep-ghost) is proposed to build the backbone of the LRID, which combines compactness and multi-branch learning. Comparative evaluations of detection performance across various models demonstrate the superiority of the LRID. The model achieves a detection speed of 130 τFPS and exhibits a significant improvement in θmAP(0.5:0.95) from 64.5 % to 68.2 % and a decrease in GFLOPs (giga floating point operations per secod) from 17.3 to 16.8, relative to the baseline. Ablation experiments further reveal that the backbone contributes a 4.8 % improvement in θmAP(0.5:0.95), and elucidate the structural design of the backbone. Ultimately, the detection accuracy for most defect classes within B-scan dataset approaches 90 %, indicating the effectiveness and feasibility of the proposed method.

Real-time human pose estimation (HPE) using convolutional neural networks (CNN) is critical for enabling machines to better understand human beings based on images and videos, and for assisting supervisors in identifying human behavior. However, CNN-based systems are susceptible to adversarial attacks, and the attacks specifically targeting HPE have received little attention. We present a gradient-based adversarial example generation method, named AdaptiveFool, which is designed to effectively perform a keypoints-invisible attack against OpenPose by aggregating the loss function of human keypoints and generating adaptive adversarial perturbations. In addition, we introduce an object-oriented perturbation generation method during the AdaptiveFool process to eliminate background perturbations. Our proposed method adapts the adversarial perturbations and generates object-oriented perturbations. On COCO 2017 datasets, our method achieves 6.3% mean average precision on OpenPose. This research provides inspiration for future work on developing efficient and effective adversarial example defense methods for HPE.

Virtual coupling, which provides great advantages in operational flexibility and line capacity, is an advanced signaling concept for the railway industry. The dynamic disturbances caused by line conditions might change the coupled movements of the train convoy into abnormal states, which means the speed difference and separation distance of the adjacent trains exceed the given thresholds. This paper proposes a method to maintain the coupled states using terminal sliding mode control (SMC) based on second-order nonlinear train dynamics, with a nonlinear observer eliminating the estimation error due to time-varying measurement delay. The controller calculates the optimal unit effective tractive force for the following train in real-time, taking the leader velocity and desired separation as control targets. The simulation of a two-train convoy on a high-speed railway is conducted including different abnormal scenarios. The results demonstrate that the proposed method eliminates the observation errors and achieves synchronous convergence of the tracking errors while guaranteeing passenger comfort, and that it outperforms traditional SMC.

As the density of railway operations increases annually, train delays typically result in a greater degree of economic loss. If we can accurately assess the current evolution of delayed trains, railway authorities will be able to make timely decisions and reduce travel anxiety among passengers. In this paper, a deep learning model integrating wavelet transform (WT), fully connected neural network (FCNN), and long short-term memory (LSTM) network is proposed to address the short-term delay prediction problem for trains. Its predictive performance is enhanced by introducing two front train selection schemes, two methods for converting train operating time features into time-series data, and five commonly used wavelet functions. For validation purposes, the model is tested using real-world operational data from two distinct high-speed railway lines in China, and common machine learning algorithms are compared, including random forest (RF), gradient boosting regression tree (GBRT), XGBoosting (XGB), and support vector regression (SVR). The final experimental results demonstrate that our proposed model outperforms the benchmark models and can be effectively applied to practical prediction problems.

The rescheduling of train timetables under a complete blockage is a challenging process, which is more difficult when timetables contain lots of trains. In this paper, a mixed integer linear programming (MILP) model is formulated to solve the problem, following the rescheduling strategy that blocked trains wait inside the stations during the disruption. When the exact end time of the disruption is known, trains at stations downstream of the blocked station can depart early. The model aims at minimizing the total delay time and the total number of delayed trains under the constraints of station capacities, activity time, overtaking rules, and rescheduling strategies. Because there are too many variables and constraints of the MILP model to be solved, a three-stage algorithm is designed to speed up the solution. Experiments are carried out on the Beijing–Guangzhou high-speed railway line from Chibibei to Guangzhounan. The original timetable contains 162 trains, including 29 cross-line trains and 133 local trains. The simulation results show that our model can handle the optimization task of the timetable rescheduling problem very well. Compared with the one-stage algorithm, the three-stage algorithm is proved to greatly improve the solving speed of the model. All instances can get a better optimized disposition timetable within 450 to 600 s, which is acceptable for practical use.

Virtual coupling, which has higher line capacity and operational flexibility, has received extensive attention because of the continuous growth of rail transport demands. The dynamic formation transition of virtual coupling train convoy is an essential foundation for improving operational flexibility. This paper proposes a transition control approach of virtual coupling train convoy formation based on model predictive control (MPC) and the formula for the reference state of trains in the coupling/decoupling scenario. Taking the reference state as the control objective, this approach outputs the driving/braking force to the power system by solving the optimal control problem during the prediction horizon based on the nonlinear train dynamic model. Since the formation transition of the virtual coupling train convoy mainly occurs near the station with complex line conditions and a significant impact on the line capacity, this paper carries on the simulation verification in the station scene. The simulation results show the effectiveness and smoothness of this approach in the formation transition of virtual coupling train convoy, while all trains meet the safety restrictions, and provide references for the decoupling position selection of train convoy during the approach to the station.

To study the driving safety characteristics of autonomous vehicles in existing transport infrastructure, this paper treats vehicle stability as a research breakthrough to study the safety characteristics of autonomous vehicles on curves and ramps by analyzing the viaduct approach context. Conducting force analysis of a vehicle in curve and ramp settings can determine the relationship between the vehicle's lateral stability on the curve and its longitudinal stability on the slope. Further, force analysis can establish the autonomous vehicle's speed and stability. This model is used to calculate the safe driving speed threshold of autonomous vehicles; their driving safety is guaranteed by controlling their speed. The calculation results show that severe weather conditions, such as rain and snow, have a significantly stronger impact on vehicle stability than factors such as vehicle models and load capacity. In fine weather, load capacity has a greater impact on the stability of heavy trucks. The safe speed threshold of autonomous vehicles, calculated in real time, is often much better than the road alignment design speed. Compared with conventional manned driving, the autonomous vehicle safe speed threshold can improve traffic efficiency to a certain extent and reduce traffic jams to enhance the traffic environment.

The Track circuit-Based Train Control(TBTC)-Communication-Based Train Control(CBTC) dual-mode onboard system is the key to realizing rail transit multi-network integration.Its mode switching is characterized by strong randomness and concurrency and directly influences the service availability of onboard signal systems.The degradation of onboard signal system faults leads to the reduction in rail transit resource utilization.The degree of reduction is reflected in the increase in train tracking time interval, which depends on the length of the section and the length of the detection section occupied by the TBTC train.From the perspective of resource allocation, use, and release under different modes, the train tracking operation scenario of the TBTC-CBTC dual-mode redundant onboard signal system is modeled using Colored Petri Network(CPN) during train operation to simulate the randomness of CBTC onboard signal system failure and the influence of system degradation on subsequent trains and accurately describe and analyze the impact on train operation.The typical configuration parameters of urban rail transit projects are substituted into the CPN model for simulation to verify the extent to which mode switching influences the operation interval under the conditions of different lines.Based on the simulation results, when the section length is at most 1 500 m, the train delay can be controlled within 180 s and the delay time will increase with an increase in the section length.

Adversarial attacks have stimulated research interests in the field of deep learning security. In terms of autonomous driving technology, instance segmentation can help autonomous vehicles identify a drivable area in a driving environment and a traffic accident will happen once instance segmentation is attacked by adversarial examples. However, only few studies have been done on adversarial attacks about instance segmentation. It would be more meaningful and more practical to research adversarial attacks against instance segmentation. In this paper, we propose the improved Projected Gradient Descent (PGD) to produce adversarial examples on the total loss of You Only Look At CoefficienT (YOLACT) instance segmentation. Firstly, we design the loss function and calculate the adversarial gradient. Then we propose the improved PGD. Finally, we obtain adversarial examples of YOLACT instance segmentation. Our attack method is efficient and powerful in both white-box and black-box attack settings, and is applicable in a variety of neural network architectures. On COCO 2017, under white-box attacks, our method achieves 1.14% box mean average precision (mAP) and 1.50% mask mAP on YOLACT with ResNet101 backbone. We also compare the similarity between clean images and adversarial examples among three different backbones and compare the operation time among three different backbones. We also find that box regression loss, classification loss and mask loss are also effective separately for generating adversarial examples. Our research will provide inspiration for further efforts in efficient and effective defense methods on instance segmentation.

为了克服多时段控制模型数据输入源连续型数据离散化选取的随意性与经验性,提出一种基于传感网与人工智能理论相结合的交叉口多时段控制深度注意力递归网络输入源选取优化方法.利用Synchro与Sumo仿真评价功能模块对数据输入源进行标准化处理.以已标记好的控制方案中起始时间等关键属性作为模型输入,同时以最优数据输入源选取点为数据输出构建模型.通过对整个模型输入层、中间层、输出层和优化方法进行仿真实现,并以某城市实际交通流量数据为测试数据进行评价对比分析.结果表明,该创新模型与传统50％位、80％位取值法相比,信号配时方案更加精准高效,交叉口全天总延误时间有效降低.

For train velocity measurement and positioning system in train control system, Doppler radar or accelerometer is usually utilized to detect the slip/slide. However, the complicated environmental conditions or the uneven rail brings the big measurement error and the poor reliability. In this paper, the train velocity measurement and positioning system based on spatial filter has been described with the advantage of simple structure, high measurement precision, high robustness, and the non-contacted measuring method. In the train velocity measurement and positioning system architecture, the wheel speed sensor and spatial filter speed sensor are integrated to detect the train speed, and the beacon system is utilized to localize the position of the train. The fault diagnosis and fusion algorithm is depicted in detailed. According to the field test results, the slip/slide can be detected accurately and reliably, and the train speed compensation can be implemented by fusing the spatial filter sensor.

“CBTC (Communication Based Train Control) + TBTC (Track-circuit Based Train Control)” is a dual-signal redundant train control system. The TBTC is the fallback mode, which is mainly implemented in the re-signaling program of the existing TBTC line. This study uses reliability block diagrams to quantitatively analyze the reliability of the dual-signal system and the time-colored Petri net of degraded operations to calculate the delay time caused by signal equipment failure leading to mode transition. These methods are applied to “CBTC + TBTC” re-signaling program of Shanghai Metro Line 2. The results show that the reliability of the new system is higher than that of the existing TBTC system and a single CBTC system, and the train delay caused by the fault leading to system fallback is minimized to 1 min. The analysis method also provides references for other multi-mode train control systems.

点式-基于通信的列车控制(BM-CBTC)多模列控系统的模式切换时延对城市轨道交通运营效率有着较大影响.分析点式-基于通信的列车控制(BM-CBTC)系统模式切换功能及过程,建立基于有色Petri网(CPN)的模式切换模型,研究不同点式(BM)控制系统设计运行间隔下的系统切换实时性.实验表明,切换时延随着BM设计运行间隔增大而增加,当BM设计运行间隔为2 min时能将列车运行晚点时间控制在5 min以内.建立的CPN模型亦可为其他配置的多模列控系统切换的实时性分析作为参考.

Accurate prediction of train delay recovery is critical for railway incident management and providing passengers with accurate journey time. In this paper, a two-stage prediction model is proposed to predict the recovery time of train primary-delay based on the real records from High-Speed Railway (HSR). In Stage 1, two models are built to study the influence of feature space and model framework on the prediction accuracy of buffer time in each section or station. It is found that explicitly inputting the attribute features of stations and sections to the model, instead of implicit simulation, will improve the prediction accuracy effectively. For validation purpose, the proposed model has been compared with several alternative models, namely, Logistic Regression (LR), Artificial Neutral Network (ANN), Support Vector Machine (SVM) and Gradient Boosting Tree (GBT). The results show that its remarkable performance is better than other schemes. Specifically, when the error is extended to 3[Formula: see text]min, the proposed model can achieve up to the accuracy of 94.63%. It proves that our method has high value in practical engineering application. Considering the delay propagation of trains is a complex process, our future study will focus on building delay propagation knowledge base and dispatcher experience knowledge base.

Automatic license plate recognition (ALPR) has made great progress, yet is still challenged by various factors in the real world, such as blurred or occluded plates, skewed camera angles, bad weather, and so on. Therefore, we propose a method that uses a cascade of object detection algorithms to accurately and speedily recognize plates’ contents. In our method, YOLOv3-Tiny, an end-to-end object detection network, is used to locate license plate areas, and YOLOv3 to recognize license plate characters. According to the type and position of the recognized characters, a logical judgment is made to obtain the license plate number. We applied our method to a truck weighing system and constructed a dataset called SM-ALPR, encapsulating pictures captured by this system. It is demonstrated by experiment and by comparison with two other methods applied to this dataset that our method can locate 99.51% of license plate areas in the images and recognize 99.02% of the characters on the plates while maintaining a higher running speed. Specifically, our method exhibits a better performance on challenging images that contain blurred plates, skewed angles, or accidental occlusion, or have been captured in bad weather or poor light, which implies its potential in more diversified practice scenarios.