Existing dedust algorithms have limitations such as low time efficiency and low image contrast enhancement affected by haze. To address these problems, this paper proposes a dedust enhancement algorithm to improve the saturation and contrast in HSI space on the basis of colour correction. First, the green channel is used to correct other colour channels in RGB colour space, achieving colour balance of sand-dust images and colour correction. Second, the adaptive saturation enhancement algorithm is designed for the S channel considering the property that all channels in HSI space do not interfere with each other, and the Contrast Limited Adaptive Histogram Equalization (CLAHE) algorithm is used to enhance the contrast for the I channel. Finally, the combined channels are transferred to RGB colour space to obtain a clear dedust image. The experimental results show that the algorithm proposed in this paper increase greatly in both the clarity and contrast recovery of image details. Moreover, the algorithm in this paper improves significantly in the time efficiency, and the average algorithm efficiency is improved by about 77% compared to the original algorithm, which guarantees the real-time performance of the algorithm.

The failure maintenance of urban rail transit bridges tends to be hysteretic. To understand the variation trend of the failure location data with time in advance, it is necessary to quickly and accurately predict the specific fault location data on the bridge. Based on the measured data of bridges, a failure prediction model for urban rail transit bridges based on random forest is proposed, and the displacement parameters of bridge structures are taken as an example. First, data preprocessing is performed, and then the Pearson coefficient is used to further screen the bridge monitoring characteristic parameters. Finally, the selected features are input into the random forest prediction model for training, and the prediction results of the structural displacement are obtained. The results show that the random forest model can predict the failure of structural displacement, the mean square error of the prediction is 0.085 1, and the coefficient of determination is 98.21%.

Aiming at the problems of poor adaptability and low detection accuracy of tunnel lining crack detection methods, a lining crack detection method based on image processing and deep learning is proposed. First, taking the on-board equipment for tunnel lining image acquisition as the research object, the shadow in the image is removed by using the improved Mask smoothing algorithm, and the seam is removed by using the seam removal method. Second, the improved VGG19 network model is constructed, the efficient segmentation of lining cracks is realized through the depth learning method, and the measurement method of crack length and width based on virtual ruler is proposed to realize the efficient and accurate detection of lining cracks. Finally, the feasibility and effectiveness of this method are verified by the actual tunnel lining detection test. The test results show that the recognition rate of crack type is high, the maximum deviation of crack length is 2.92 mm, and the maximum deviation of crack width is 0.28 mm.

Many existing neural network-based depth map denoising methods ignore the difference between depth map structure information and detail information, leading to the fact that these methods often fail to accurately recover the geometric structure information of the depth map. To this end, this paper proposes a single depth map denoising method that combines image high-frequency (HF) and low-frequency (LF) decomposition and multi-scale two-level fusion (MTF) strategies. First, considering the difference between different noises, a multi-scale Gaussian filter is introduced to decompose the depth map containing noise into a group of LF structural components and a group of HF detail components. Secondly, considering the complementary characteristics of LF structure information and HF detail information, these two groups of components are input to the LF feature extraction network (LFN) and HF feature extraction network (HFN) based on MTF, and a multi-stage feature fusion and feedback is proposed using a complementary feature weighting fusion mechanism between these two networks. Finally, the HF and LF enhanced features output from the HFN and LFN are predicted using the high-and low-frequency merged reconstruction module for residuals, and then fused with the input map to obtain a high-quality depth map. The experimental results show that the proposed method has better performance than several mainstream deep graph denoising methods such as ARCNN, Fast ARCNN, DnCNN, ADNet, and FFDNet in terms of peak signal-to-noise ratio, root mean square error, structural similarity, and comprehensive performance comparison.

Most of the existing combined image retrieval algorithms directly perform feature joint embedding to obtain fused query information, ignoring the complementary information between different modalities and retaining a large amount of redundant information. To address these issues, a combined image retrieval algorithm based on adaptive and cascade structure is proposed. First, the proposed algorithm uses the adaptive bilinear pooling module to fuse the information between multi-modalities, then adopts the adaptive mechanism to realize the filtering of information in each modality. Second, the cascade structure is used to explore the correlation among multimodal information at both local-global levels and fuse the valid information to generate the final embedding vector for combined query. The experimental results show that the algorithm accurately characterizes the cross-modal combined information and improves the retrieval accuracy on multiple datasets.

In view of the platoon longitudinal control of unmanned vehicle platoon driving on the road with low adhesion coefficient, the calculation method of the minimum safe distance, considering the road adhesion coefficient and the acceleration information of the front vehicle failing to be continuously obtained, is proposed as an improvement to the traditional Constant Time Gap (CTG) algorithm for the desired vehicle distance. Then, a hierarchical structure is used to design the platoon control system, and the platoon stability is analyzed to derive the conditions for the controller gain in the upper-layer control. Finally, a platoon of four vehicles is simulated and verified. The results show that string stability of the vehicle platoon can be guaranteed and collision between two cars can be avoided by the improved constant time gap algorithm when the road adhesion coefficient is low. In the case that the car cannot obtain the acceleration information of the front car and the front car brakes urgently, the maximum deviation rate of the platoon with the improved constant time interval algorithm is 19% during the driving process, which is lower than the maximum deviation rate of 22.27% when the traditional method is applied, and the distance error of the platoon eventually converges to zero with the increase of time. The maximum value of vehicle acceleration in the driving process of the platoon that uses the improved constant time interval algorithm is 0.75?m/s2, which is lower than the maximum value of acceleration 0.91?m/s2 using the traditional method. It shows that the stability of the platoon and higher driving comfort can be guaranteed by the improved constant time interval algorithm.

With the development of air transport industry, efficient and accurate demand forecasting has become an important condition for civil aviation units to improve operation efficiency and enterprise competitiveness. This paper considers the spatio-temporal characteristics and figure structural features of passenger demand in Air Transportation and constructs the Temporal Graph Convolutional Network model (AT-TGCN) based on Graph Convolution neural Network (GCN), Gated Recurrent Neural (GRU) and Attentional mechanism for the small network of transit routes. First of all, this paper builds a small network of transit routes based on the historical flight operation data of airlines. Secondly, multidimensional features of time and space are analyzed according to the topology of transit route network. Then, a prediction model is constructed based on the spatio-temporal characteristics of the network, and the prediction research is carried out based on specific data. Finally, we compare the prediction model with the traditional model by multiple error indicators. The prediction results show that the prediction model based on spatio temporal graph convolution has significant advantages in passenger demand research of transit airline network and the prediction accuracy can reach 93.10%, which provides a valuable prediction model for airlines to understand the demand changes of transit passengers.

Fisheye cameras can capture scenes with a large field of view, but they introduce severe radial distortion in the image. Most existing deep learning-based methods require corresponding calibration labels such as distortion parameters and distortion-free images, etc., while the above assumptions and requirements may not hold when applied to real-world scenarios. Therefore, this paper proposes a self-calibration radial distortion correction algorithm based on generative adversarial networks in the polar coordinate domain. This is to achieve self-calibration fisheye correction of a single image by exploiting the horizontal consistency of distortion size in polar coordinate images and the smoothness of distortion within polar coordinate images. The polynomial model and the division model are used to synthesize a large range of distorted training datasets so that the proposed model obtains a better generalization capability. The comparison experimental results on the synthetic and real datasets demonstrate the superiority of the method.

This paper proposes a new end-to-end decision-making scheme for lane keeping based on the improved TD3 algorithm. First, a multi-data fusion TD3 algorithm framework is constructed to perceive the kinematics data information and visual image information of autonomous vehicle to enhance the stability of the algorithm. Second, combined with the concept of attention mechanism, image features are refined so that the algorithm pays attention to the key road information, which enhances the interpretability of the algorithm. Then, a guidance-based reward function is designed with comprehensive consideration of driving safety, comfort and efficiency to guide the intelligent agent to learn a more human-like driving strategy. Finally, a classification and high-value prioritized experience replay method is applied to improve the sample utilization and accelerate the algorithm convergence. With the aid of TORCS simulation platform, multiple sets of comparative experiments are designed to verify the effectiveness and feasibility of the proposed method. Furthermore, through simulation tests in multiple scenarios, it is verified that the overall performance of the proposed improved TD3 algorithm is better than that of the TD3 algorithm.

With the rapid development of artificial intelligence, intelligent inspection technology gradually replaces manual inspection, effectively improves inspection efficiency, reduces labor cost and becomes an important part of intelligent inspection system. At present, the existing intelligent inspection technology can identify fewer types of equipment components in complex scenarios and the detection accuracy and speed are relatively low. Small parts (such as buttons, etc.) have poor recognition performance. In order to achieve better detection effect and a certain degree of wide applicability, an intelligent robot inspection and identification algorithm based on improved SOLOv2 is proposed. This method can better deal with the problem of accurate positioning and modal extraction of various types of equipment components in complex scenes and solves the problem that the existing intelligent inspection technology can identify few types of components. Aiming at the problem that the instance segmentation algorithm SOLOv2 has low recognition accuracy for small objects, by adding the output of large-size hierarchical feature maps in the feature pyramid network, the number of positive samples of small target objects is increased and the recognition accuracy of small objects is improved. The experimental results show that the method proposed in this paper has better recognition accuracy and higher robustness in complex scenes than the current inspection and identification algorithm; compared with the original intelligent inspection system, the number of identifiable components increases 12 categories; compared with the original SOLOv2 algorithm, the accuracy of small target objects is improved by about 10% and the overall recognition accuracy is also improved by 1.7%.

The GNSS-based train positioning system reduces the cost of railway construction and operation. And the correct detection of the turnout section where the train is located is a prerequisite for train satellite positioning. To improve the accuracy and real-time performance of the train track occupancy detection process, a Particle Filter (PF)-based train track occupancy detection method is proposed. First, based on the PF principle, an adjustable error confidence zone is set to retrieve the adjacent turnout nodes according to the current position of the train, effectively avoiding the problems of missed retrieval and repeated retrieval. Second, the track occupancy detection process is modeled as a real-time matching process between the train running position and the candidate sections of the route map. And particles are randomly generated on the candidate turnout section connected with the train to track the train in real time. The distance error and heading angle error between the particle motion state and the train positioning point at each moment are used as weighting functions. By calculating the cumulative particle weights of the track sections adjacent to the turnout nodes, the maximum probability of occupying the section is derived to achieve the correct detection of the turnout sections where the train is located. Finally, the measured data is compared with the existing track occupancy detection method. The experimental results show that the PF-based train track occupation detection method can effectively identify the current track. And its average detection distance is stable and is reduced by about 6.5 m compared with the direct projection method and reduced by 3.8 m compared with the weighted recognition method, with great real-time performance.

In the automatic operation of high-speed train, the application of traditional active disturbance rejection control to track the running speed has the following disadvantages such as too many adjustable parameters, difficult parameter setting, low tracking accuracy and weak anti-interference ability. To solve these problems, a design scheme of speed controller based on Linear Active Disturbance Rejection Control (LADRC) is proposed. First, based on the single-point train model, the linear extended state observer of the unmodeled dynamics in the real-time estimation system is designed according to LADRC. Second, the Sliding Mode Control (SMC) with improved exponential convergence law is used to optimize the nonlinear error feedback control law to improve the response speed of the system and reduce chattering. Finally, the Sliding Mode Linear Active Disturbance Rejection Control (SM-LADRC) controller is designed to track the speed command, and the effectiveness of the controller is verified by numerical simulation. The experimental results show that SM-LADRC controller has the advantages of reduced adjustable parameters and easy adjustment, high tracking accuracy, strong disturbance rejection ability and fast response speed.

Aiming at the problem that the traditional combinatorial test generates redundant test cases covering non-interactive parameter combinations in the CBTC train control vehicle TSM monitoring curve, this paper proposes an automatic reduction method for combinatorial test case suites based on decision tree equivalence. Firstly, combine the combinatorial test cases covered by t-way parameters and their output to construct a data set that captures System Under Test (SUT) behavior, and use the CART algorithm to infer the data set into a decision tree; Secondly, an improved reduction algorithm for combinatorial test cases is designed using the equivalence of decision tree structure and misclassification equivalence to reduce redundant combinatorial test cases; Finally, the reduction algorithm is used to analyze the function of the CBTC level train control vehicle TSM monitoring curve. The relevant experimental results show that by using this method, the reduction rate of 74% can be achieved. At the same time, the low-level combination coverage and fault detection capabilities of the test suite are basically the same before and after the reduction.

The division of block sections is one of the most important influencing factors of the design capacity. And it is also closely related to the cost of traction power supply equipment, which is critical to guide the economic, reasonable and scientific system design and planning of the ultra-high speed maglev trains. In this paper, the average headway and the project cost of block sections are seen as the objectives. By constructing the capacity computing model of ultra-high-speed maglev train, the problem of dividing block sections is transformed into a mixed-integer nonlinear programming problem, whose design is optimized by a Non-dominated Sorting Genetic Algorithm Ⅱ (NSGA-Ⅱ). Based on the data in ultra-high-speed maglev trains, including those regarding train lines, vehicle and signal system, a simulation case is designed to evaluate the above algorithm. The simulation results are then compared with that of the Vector-Evaluated Genetic Algorithm (VEGA) based on the linear weighting method. The results show that the quality and speed of the Pareto optimal solution generated by NSGA-II are better than those by VEGA, which contributes to satisfying the constraints accurately.

Hybrid Electric Vehicle (HEV) is usually driven by two or more power sources, and the power distribution of each power source is coordinated through control algorithms to improve the vehicle performance and vehicle fuel economy. This paper takes the target power-split HEV as the research object. A “black box” modeling approach is used to evaluate the influence of input variables on the controlled hybrid HEV and its operation modes by combining parameter identification, physical models of vehicle dynamics and drivetrain, and real-world test data. This paper proposes a two-layer Model Predictive Control (MPC) energy management strategy. Taking advantage of Vehicle-to-everything (V2X) information, including road speed limit, car-following constraints, intersection signal status, and stop line locations, the upper-layer controller calculates the reference operating states of the vehicle and feed these information to the lower-layer controller. The lower-layer controller calculates the torques and braking force of all components through MPC to optimize the operation process of the HEV. Based on the simulation in matlab/Simulink, the feasibility and effectiveness of the proposed method is verified, and the state of charge (SOC) consumption of the battery is reduced by 5%.

In the variable condition bearing fault diagnosis task, the domain adaptation method only aligns two domains globally without considering category level alignment. To solve the above problem, a two-level confusion adversarial domain adaptation network is proposed. The network consists of a feature generator, two task classifiers, and an auxiliary classifier. We use source domain samples to assist two task classifier learning. Meanwhile, a novel adversarial learning objective based on two-level domain confusion strategy is implemented on the auxiliary classifier. Through adversarial training, the features generated by the feature generator can align at category point. The results on two bearing datasets show that the diagnostic accuracy of our method is higher than the traditional deep learning algorithm and the other four domain adaptation algorithms.

Stray current is an important hidden danger that affects the operational reliability of urban rail transit power supply system. In this paper, a dynamic stray current distribution model is established and its dynamic distribution characteristic is simulated by taking the rail-drainage net-ground model as the research object under the condition of considering the additional resistance of train operation. The research results show that the peak value of stray current mainly appears in the train starting and braking stage, and the peak value is larger in the braking stage. Compared with the fast traction condition and the comfortable traction condition, the stray current generated in the economic traction condition is the smallest. After considering the multiple additional resistance, the stray current increases by 14.65% at the starting stage of economic traction condition and increases by 7.09% at the starting stage of fast traction condition, and it is almost unchanged in the braking stage of fast traction condition and the comfortable traction condition. At the same time, the stray current decreases by 6.45% at the braking stage of economic traction condition. Related conclusions provide more accurate references for the design and protection of stray current drainage devices for urban rail transit.

To address the magnetic core saturation caused by unbalanced traction currents in 25 Hz phase-sensitive track circuits, a modeling method for the nonlinear model of a choke transformer is proposed. The approach combines the JA (Jiles-Atherton) hysteresis model and the three-capacitor transformer model. Firstly, a JA hysteresis model parameter identification algorithm based on Particle Swarm Optimization (PSO) is designed. This algorithm, along with the energy calculation method, is utilized to extract the JA hysteresis loop parameters of the iron core and the distribution parameters in the nonlinear model of the choke transformer, respectively. Subsequently, by comparing the calculated distribution parameters with Maxwell's simulation results, it is concluded that the parameters, such as leakage inductance and distributed capacitance of the choke transformer, extracted by our method exhibit good accuracy. Finally, the results of solving the BH curve, excitation current, and output voltage curves of the nonlinear choke transformer model are compared and verified against Pspice simulation results. The research results indicate that the proposed model can be utilized to accurately analyze the saturation problem of magnetic cores in choke transformers. The saturation state of the magnetic core can be determined by analyzing the real-time hysteresis loop shape of the model core or the distortion of the primary current in the choke transformer.

The continuous extension trend of the mileage between the first-level maintenance of Electric Multiple Units (EMUs) makes multi-day EMU circulation an important organizational form to improve the efficiency of EMU operation. This paper takes the problem of EMU circulation allowing multi-day itineraries under the unfixed routing mode as the research object. Based on the one-day train operation diagram, this paper defines a network with train operation lines as nodes and the connecting relations between train operation lines as arcs, so that the EMU circulation problem is converted into a network model. On this basis, the expansion of the multi-day train operation diagram is realized by defining the succession time on terminal arcs, and the optimization of train itineraries consisting of the succession overnight corresponding to the shortest path on the network of the expanded train operation diagram, which is convenient to solve using a general algorithm. A case study on an intercity railway line shows that the method proposed in this paper can flexibly generate EMU circulation plans for either one-day itineraries only or multi-day itineraries allowed. The plan of two-day itineraries generated is better than the existing plan. Under the same traffic of this intercity line, the efficiency index of EMU is improved from 23.22%, when only one-day itineraries are allowed, to 29.03% and 38.71%, when two-day and three-day itineraries are allowed, respectively.

This paper investigates the congestion prediction and resolution methods in railway network, in order to accurately predict the railway congestion based on the actual traffic flow data, and to meet the transport capacity of the network while effectively relieving traffic congestion. The dynamic traffic assignment of railway flow is realized based on the actual railway flow paths, comparing real-time traffic flow with transport capacity to predict the traffic congestion. The discrete space-time network is constructed based on the given railway network to represent the railcar flow organization. To minimize the transportation mileage and maximize the completion rate of transportation plan, a mixed-integer programming model is built. The model considers the constraints of train reorganization operation, real-time storage capacity of stations, and maximum train formation capacity of sections. The model optimizes the traffic flow organization operation, so as to realize the effective resolution of railway traffic based on the given physical network of railway transportation. A railway transportation network with 19 stations and 22 sections is selected as for the numerical experiment. Four test cases with different OD traffic sizes are designed and solved by Gurobi, a mathematical programming optimizer. The optimal solution of all 4 test cases can be obtained within 1 hour. Time and duration of the railway congestion are identified based on the results of case 4, the changes in capacity utilization of railway network are compared, and a sensitivity analysis is performed on the time interval parameter. The results show that the proposed model can effectively solve the congestion of railway traffic.

In the context of building a multi-level rail transit network within the urban agglomeration, the integration of interchange services at interchange hubs has not yet been realized, which is a bottleneck restricting the merging of multi-standard rail transit networks. To solve the problems of large number of people gathering on the platform and waiting for a long time at single-platform-interchange between urban railway and subway, this paper proposes an optimized method of coordinating railway timetable in multi-standard rail transit lines. Firstly, this paper analyzes the characteristics of passenger transfer at single-platform-interchange connecting urban railway and subway and explores the current passenger travel pattern under ticket-transfer situation. Then, from the perspectives of station safety and passenger satisfaction, an optimized model for railway timetable coordination is established for the interchange station. The optimization objectives are to achieve the minimum number of people gathering on the platform and the shortest average passenger waiting time considering uneven departure interval. Finally, taking Chengdu Xipu Station as an example, which is the first single-platform-interchange station between urban railway and subway in China, the validity of the model is verified. The research results show that the optimized timetable for the interchange subway and railway can significantly reduce the maximum number of people gathering on the platform and the waiting time. The former decreases by 43.6% and the latter is reduced by 26.2%.

In the scenario where the intersection is congested and there is a large difference between the main road and the slip road traffic, the policy for automated vehicles, the First Come First Serve (FCFS) policy, usually has lower efficiency than that of the signal timing policy. To solve that paradox, a hybrid control method for unsignalized intersections is proposed, which combines the idea of FCFS policy and platoon-based policy. The proposed hybrid control method optimizes vehicles’ service sequences to minimize the overall delay while ensuring safety. By introducing a heuristic scheduling policy and cancellation reservation mechanism, vehicles’ priorities are distinguished according to the traffic flow in corresponding lanes. Vehicles are adaptively formed into platoons through the intersection according to the real-time traffic flow conditions, thus reducing the frequent right-of-way exchanges generated by the FCFS strategy and ensuring the continuity of vehicle traffic on arterial roads. Simulation results show that in the paradox scenario, the hybrid control policy can reduce the total delay by 55.84% compared with the FCFS policy. When the overall traffic volume at the intersection is high and the volume difference between the arterial and branch is significant, the hybrid control policy has advantages in reducing delay and improving intersection capacity than the FCFS policy.

With the rapid development of Internet of Vehicles technology, Cooperative Adaptive Cruise Control (CACC) has become the main trend of future smart transportation development. Based on the Gipps safety distance algorithm, a dynamic safety distance is introduced to consider the driving characteristics of CACC vehicles and traditional vehicles, and a refined numerical simulation model of mixed traffic flow is proposed. Combined with the evolution rules of small-scale grid cellular automata, the basic mixed traffic flow characteristics of different vehicle ratios under different reaction times are compared. The results show that the basic features of the mixed traffic flow considering the influence of the dynamic safety distance agree with the characteristics of the three-phase traffic flow theory. Meantime, increasing the proportion of CACC vehicles and reducing the reaction time can increase the average velocity of vehicles and improve the overall traffic congestion and flow in the lane. And the improvement effect is more significant under vehicle congestion.

Based on the operational mechanism of the travel reservation strategy, the cumulative prospect theory is applied to analyze the travel mode choice of travelers under uncertain environment after the implementation of the strategy. It discusses the feasibility from two aspects: individual utility and social welfare. The research results show that: in terms of individual utility, the travel reservation strategy can weaken the influence of individual randomness through the system spatial-temporal adjustment, significantly improve the travel efficiency of travelers, and increase the travel rate in public transportation, and the implementation effect is better in the morning peak hours than the evening peak hours. In terms of social welfare, the total social welfare brought by the strategy is higher than that of the congestion charging strategy. Therefore, compared with the traditional traffic control strategy, the travel reservation strategy is more flexible and refined, and has a greater potential to alleviate traffic congestion.

This paper aims to explore the impact of residents’ personal attributes, household attributes, travel characteristics, pandemic perception attributes, behavioral intentions, choice attitudes and other factors on travel mode choices in different stages of the COVID-19 pandemic. A mixed Logit model is constructed based on the travel data from the sampling survey of Beijing residents under three stages of the pandemic, the initial phase, the outbreak phase, and the stabilization phase. The results show that travel distance is positively correlated with travel mode choice in all three stages of the pandemic. The maximum predicted marginal values of each stage and their corresponding variables are 3.299 (5~10 km), 2.983 (>10 km), and 3.148 (5~10 km), respectively, and long-distance travel has the greatest impact on the travel mode choice. The perceived attributes of the pandemic and travel distance have obvious moderating effects on the travel mode choice. During the pandemic outbreak period, the perceived pandemic attributes, behavioral intentions, and choice attitude variables have a significant negative correlation with the travel mode choice, and residents’ psychological concern of being cross-infected during travel is obvious; only 18.8% of travelers choose to travel by bus or subway, and the travel structure changes significantly. During the stable period of the pandemic, the choice of attitude variable has a positive and significant impact on the choice of transportation mode, and the degree of influence becomes larger, indicating that travelers demand a higher rate of population contact and strictness of pandemic prevention measures for transportation mode. The research results can provide a reference basis for travel decisions of travelers under public health events and the prevention and control of the pandemic by relevant management departments.

Shared parking is a way to improve the utilization of parking resources by utilizing the staggered parking characteristics of the supply and demand ends.However, supply cancellation often occurs in real-world shared parking. If the platform does not take effective management measures, some parking users will not be able to park under the original allocation scheme. To address the parking allocation issue in the case of supply cancellation, a reallocation strategy is proposed by analyzing the impact of supply cancellation on users and the platform.Then, based on the reallocation strategy, a allocation model for shared parking space is established with the objective of maximizing the benefit of the platform after comprehensively considering user preferences. Platform revenue, parking space utilization, and user satisfaction are used for evaluation. Finally, the proposed model is validated by arithmetic experiments. The results showed that compared with the conventional scheme, the allocation scheme for the shared parking platform considering reallocation can improve the platform revenue by 23.73%, the parking utilization rate by 18.45%, and the user satisfaction by 30.61%. This has a reference value for further research on the operation and management methods of the platform.

In order to efficiently deploy approaching and departing aircraft and obtain their best sequencing, machine learning methods are used to predict the flight time of the approaching aircraft in the terminal area. The flight characteristics of the aircraft in the terminal area and the factors affecting the flight time of the approaching aircraft are analyzed. Then, 22 important features that affect the flight time prediction are proposed. The density clustering DBSCAN method is introduced to cluster different path categories of traffic flow. This paper establishes a flight time prediction model based on the integrated machine learning algorithm, XGBoost. Taking the Kunming terminal area in Yunnan province as an example, the model is trained, verified and tested, and the mean relative error and mean square error are used as evaluation indicators to analyze the error of the prediction results. The results show that compared with linear regression, support vector machine regression and artificial neural network methods, the proposed model has the best prediction results for flight time, with a prediction accuracy of 95.18% within 5 min.

As the service time of high-speed railway gradually increases, some diseases in the concrete structure of ballastless track have been accumulated and appeared gradually. Therefore, how to effectively detect these diseases has become one of the important measures to ensure the safe and effective operation of high-speed trains. After systematically combing the disease types and causes of ballastless track concrete structure, this paper analyzes the application and applicability of the current typical non-destructive testing technology in ballastless track. Besides, some future new technologies of track disease detection are introduced. Through analysis, ultrasonic technology has higher accuracy in detecting shallow cracks or cavities without impurity. For deeper cracks, the impact echo method can be used. Infrared thermal imaging technology has significant effect in predicting the specific location and evolution trend of shallow surface cracks in track slab. Geological radar technology is widely used in detecting the concrete structure of the track or the void of the mortar layer with larger internal diseases. In terms of new technology for track structure disease detection, 3D laser technology can automatically obtain high-precision disease information of concrete structure of ballastless track. Image recognition technology such as fiber optic grating and machine vision can obtain data information of ballastless track injury and damage in time to minimize the loss. In addition, based on the difference of vibration response and stiffness characteristics under train load dynamics, the track structure diseases can be identified, monitored and evaluated at non-skylight times. At last, the coordination of multiple detection technologies or multi-source characteristics fusion can further improve the identification accuracy and detection efficiency of the concrete structure diseases of ballastless track.

Based on the Feifengshan Tunnel in Hangzhou-Shaoxing-Taizhou high-speed railway, the mechanical behavior of the primary support of tunnel in the typical diatomite stratum section is studied by field test. The test items mainly include the contact pressure between the surrounding rock and primary support and that between the primary support and secondary lining, the stress of steel rib or lattice girders, and the strain of shotcrete. The test results show that the pipe shed is applicable to the diatomite stratum and can effectively control the vault settlement at the tunnel portal section. The advanced small pipe grouting is applicable to the weak diatomite stratum with short self-stabilization time in the middle of the tunnel and can improve the surrounding rock to control the vault settlement. The safety factor of primary support gradually decreases and tends to stabilize as the tunnel construction progresses, and this process can be divided into sharp reduction stage (palm face passes within 0~20 days or 0~20 meters after working face passeal), slow reduction stage (within 20~40 days or 20~30 meters after working face passeal) and stabilization stage (after 40 days or 30 meters after working face passeal). Both steel rib and lattice girders can be used in diatomite stratum tunnel, but steel rib has stronger ability to control the vault settlement, especially to control large deformation in the early stage of tunnel excavation. The contact pressure between the steel rib and surrounding rock basically reaches its peak within 10 days, while the lattice girders reach its peak in 30~40 days before stabilizing. The relevant research results can provide reference for the similar projects in the future.

Grouting is a common method to actively control tunnel deformation. However, there is much research on grouting for tunnel lifting, but few studies on systematic mechanism and strategy of grouting control of tunnel horizontal deformation. To solve this problem, this paper proposes a comprehensive evaluation system of grouting effect based on the recovery impact of existing tunnel deformation under grouting and the impact of grouting on excavation support system. Then the paper conducts a study based on a project in Tianjin using 3D finite element numerical simulation. The results show that the recovery rate of tunnel deformation decreases rapidly with the increase of grouting distance, but when the grouting distance is less than 3 m, the grouting tends to over squeeze the tunnel, resulting in the increase of tunnel ellipticity. It is suggested to set the grouting distance at about 3 m. In order to increase the recovery rate of the tunnel ellipticity, when the tunnel is in the uplift area and transition area, the center point of the vertical grouting body should be higher than the horizontal centerline of the tunnel. In the settlement area, it should be flush with the center of the tunnel. The law of vertical deformation of tunnel caused by grouting is determined by squeezing and pushing, and the squeezing has a significantly greater impact than the pushing. Therefore, the top and bottom of the tunnel show uplift and subsidence deformation, respectively, after grouting. When the tunnel diameter is about 6 m, the length of grouting body is recommended to be 5~7 m. When using grouting to restore the deformation of existing tunnel, the influence of grouting on the deformation and internal force of foundation pit support structure should be verified and monitored.

The integrity and bearing capacity of shield tunnel structure are closely related to the mechanical properties of segment joint. To clarify the mechanical characteristics of the whole bearing process of the bolted circumferential joint, a theoretical analytical model of the segmental force in the joints is established to describe bearing behaviors of the joint under the compressive bending load. In the stage Ⅳ of joint stressing, the influence coefficient α on the height of the concrete pressure zone at the outer edge is introduced. Through orthogonal analysis and numerical simulation, it is clear that the main influencing factor of α is the joint gap, and the calculation formula of α when the outer edge concrete is crushed is further clarified. The equation shows that for most subway shield tunnels in China, α can be approximately as 0.2. Finally, the rationality of the joint segmented mechanical model is verified by numerical analysis. The results show that based on the 4 stages of the longitudinal joint bearing process and the theoretical formula of the height influence coefficient in the outer edge concrete pressure zone, a theoretical analytical model can be proposed to calculate the whole longitudinal joint bearing process of the bolted circumferential joints. The model can be used to analyze the mechanical state of segment joint under any load, which can provide relevant theoretical basis for the design of segment joint and disease prediction.

To investigate the effect of repeated unloading and loading on the long-term stability of rock masse with incrementality, granite is tested by uniaxial incremental repeated loading and unloading combined with acoustic emission monitoring. The experimental results show that under incremental repeated loading and unloading, the granite is mainly in the form of cleavage failure, and its fragmentation degree is increased compared with the conventional uniaxial loading test. Compared with the conventional uniaxial loading tests, the failure strength of granite is reduced by 12%, and the failure strength of granite under load retention is reduced by 18%. During the load retention phase, the granite still produces axial deformation, and as the load increases, the deformation generated during the load retention phase rises, and this deformation will take a longer development time to stabilize. When the stress level is between 28% and 64% of the peak stress, the Felicity Ratio (FR) is greater than 1, and the Kaiser effect is established. The fracture damage stress of granite is judged by the change of FR between the peak stress of 57% and 64%, which can obtain the precursor information of rock failure and provide reference for early warning of rock failure.

Snowdrift is a common weather phenomenon in cold areas, which can reduce the traffic visibility on expressway and affect driving safety. This paper summarized the influencing factors of snowdrift and its relationship with subgrade structure based on the field investigation and environmental monitoring of snowdrift hazard-prone areas along the Beijing-Xinjiang Expressway. The multiphase flow numerical model is established based on the measured parameters of snow particles, and the effect of different structure parameters of the snow guard is studied. The results show that the complex topography and climate conditions lead to huge spatial-temporal heterogeneity of snow distribution due to snow drifting. The snow is easier to be deposited on the road cutting form than the embankment form, and the snow platform and snow guard can significantly reduce the snow on the road surface. The gap at the bottom of the snow guard accelerates the flow field near the ground and the snow particles extend to the leeward side, the height of the snowboard will affect the snow deposition range on the leeward side of the snow guard, and the porosity will affect the snow accumulation pattern in front of and behind the snow guard. For areas with heavy snowfall and large snow storage space, the gap at the bottom and the height of the snow guard can be appropriately increased, and it is recommended that the porosity of the snow guard be set between 0.4 and 0.6. The research results can provide technical basis for the prevention and control of snowdrift hazard on expressway.

With the development of urban rail transit express and suburban railway lines, the operating train speed can reach 160 km/h. The increase in train speed can affect the vibration damping performance and dynamic responses of the Floating Slab Track (FST). To investigate the vibration and noise reduction of steel-spring FST under different train speeds, a 200 m test section of steel-spring FST is designed and constructed in the circle test line of the National Railway Test Center. Three test sections of steel-spring FST and regular slab track are selected to carry out the experimental research on vibration and noise reduction effect and dynamic performance of track under different train speeds. During the test, the train speed increases from 60 to 160 km/h step by step. The results indicate that with the increase of train speed, the displacement of FST gradually decreases, the displacement of outer rail gradually increases, the displacement of inner rail gradually decreases, and the displacement of rail and slab meet the specification requirements under all train speeds. The displacement of a “2/3 arrangement” steel-spring FST is greater than that of the “2/2 arrangement” one. With the increase of train speed, the vibration damping effect of steel-spring FST increases first and then decreases slightly. When the speed is larger than 100 km/h, the vibration damping effect is higher than 10 dB. Compared with regular slab track, a noise reduction effect with the maximum value of 3.7 dB(A) can be obtained by using FST.

This paper aims to improve the feasibility and technical advances of the engineering application for the wind resistance braking device of the 400 km/h high-speed train and address the problem of insufficient braking force in the high-speed stage of the train. Based on the N-S and k-ε double equation models of three-dimensional steady incompressible viscous flow field, taking the streamlined appearance of CR400AF platform EMUs as the prototype, a new type of “butterfly” wind resistance braking device is assembled, and the hydrodynamic calculation models under different working conditions are established. The aerodynamic characteristics of high-speed trains with and without brake windshields at different speed levels are compared and analyzed. Then, in the 400 km/h high-speed train under the action of wind load during emergency braking, the strength and vibration characteristics of the first-row brake wing plate are calculated and analyzed. The results show that the air resistance of the high-speed train with the first-row brake flaps has increased significantly compared with that without the first row of brake flaps under different speed levels. And the drag coefficient of the whole train increases from 0.234 to 0.255. When the first-row 42Cr4 material is employed in the 400 km/h high-speed train, the air resistance of the whole vehicle increases by about 14.0%, and the aerodynamic load on the first-row brake air wings is about 3.72 kN. The maximum stress point under the action of the wing plate occurs at the connection of upper pull rods on both sides of the air wings, which is far lower than the material yield limit, meeting the requirements for strength design. The inherent frequency range is 143.53~845.64 Hz for the first 12 order modal vibration in front of the first row of brake air wing plate in high-speed train, in which the first to fifth order main mode is the bending and torsional vibration in the upper part of the air wing plate, and the sixth to twelfth order gradually extends to the bending and torsional vibration in the lower part and both sides, accompanied by local modes.

To explore the changes in stiffness characteristics of the axle box rubber joint and its influence on fatigue life for CRH3 high-speed EMU, the rubber joint stiffness is tested after 1.2 million kilometers of service mileage, and the distribution characteristics of stiffness are obtained through the hypothesis test. The Ogden constitutive of rubber is fitted to establish an accurate finite element model for rubber joint based on the uniaxial and biaxial test data. On this basis, the stress, strain, displacement, and the changing law of the stiffness are analyzed. The fatigue life for the joint under cyclic load is obtained by using the principal stress method and the S-N curve according to the hyper-elastic properties of rubber. The results show that the radial stiffness and the axial stiffness obey normal distribution N (105.12, 7.42), N (12.95, 1.752), respectively, for the rubber joint in service of 1.2 million kilometers. The radial stiffness and axial stiffness fluctuate and decrease with the increase of load, and the stiffness changes have obvious nonlinear characteristics. As the load increases, the fatigue damage area gradually expands, and the fatigue life gradually decreases. The minimum fatigue life is about 1.14 million times. The research provides a reference basis for the design and optimization of rubber joint.

To improve the transportation efficiency of China Railway Express (CRE) and reduce transportation costs, according to the current situation of stations and lines in the transportation network, a hybrid hub-and-spoke CRE transportation network is constructed. On this basis, the time factor is added to form a dynamic transportation service network for the CRE. Combined with the characteristics of the dynamic transportation service network, the four-dimensional decision variables are set, and the optimization model of CRE operation plan is constructed with the constraints of freight flow balance, train succession, marshalling capacity limitation, train service capacity limitation and transportation time limitation, and the lowest comprehensive operation cost as the optimization target. The transportation network composed of 15 domestic sources and 7 overseas demand areas is selected as an example for analysis and verification. The results show that when the transportation demand is 17, the operation plan includes 9 assembly trains and 8 trunk trains. Compared with the fully direct operation plan, the “direct+transit” operation plan obtained by the model can reduce the overlap rate of CRE transportation lines and save 37 hours of total time consumption when transporting the same goods 347 FEU. The optimized operation plan of CRE can reduce transportation time and improve transportation efficiency, which provides references for the formulation of daily operation plan of CRE.

The real-time dynamic coupling under the virtual-coupling technology provides a guarantee for running multi-group trains in different sections based on the spatial and temporal distribution characteristics of passenger flow. After analyzing the train operation plan based on full-length and short-turn routing under the virtual-coupling technology, this paper builds an urban rail transit capacity allocation optimization model with the objective of minimizing the train operation mileage considering the uneven passenger flow distribution in urban rail transit. And the factors considered in the model include the departure interval, the full-load rate at large and small interchange sections, the number of available vehicles, and the line capacity. Finally, taking Line 4-Daxing and Line 10 in Beijing Municipal Railway as examples, the optimization scheme of the two lines during the off-peak period is analyzed. The results indicate that the virtual-coupling technology combined with the operation of full-length and short-turn routing can meet the demand for capacity matching of different interchange sections and achieve a better balance of full-load rate and departure interval, which is conducive to improving passenger travel efficiency and reducing operating costs.

In order to reduce the passenger waiting time and improve passenger travel efficiency, under the condition of reaching a certain level of seat utilization, this paper proposes an optimization model for the multi-timeslot operation plan of intercity train. The arrival of passengers at the station is considered as a Poisson process, and the waiting of passengers at each station is regarded as an M/M/1 queuing problem. The Lyapunov drift is introduced to represent passenger queue changes. The optimization model is established with the goal of minimizing the average accumulated passenger waiting time and the train operation cost. Through Lyapunov drift penalties, the goal is transformed into a coordinated optimization of the train operation cost and passenger queue length, and the balance between them is achieved through setting the coefficient value of the Lyapunov drift penalty weight. Finally, the Little’s Law is used to solve the passenger waiting time in queue. Taking the Cheng-Sui-Yu high-speed rail as an example, an Improved Harmony Search (IHS) algorithm is adopted to solve the optimized operation plan. The results show that the optimized train operation plan can effectively reduce the queue length of passengers in OD pairs, the queue length of passengers in each timeslot is more balanced, the average waiting time is reduced by 28.3%, and the transportation network remains stable.

To optimize the high-speed train stop schedule, this paper considers the characteristics of driving according to passenger flow and priority of large passenger flow in railway transportation. And according to the passenger’s choice of train, that is, passengers will choose the train with the shortest transit time, a passenger flow distribution method suitable for railways is proposed and named as distribution with large passenger flow as priority. The goal is to minimize the total number of train stops, the train stop cost, and the passenger time loss, and to meet the multiple constrains including passenger flow, train transportation capacity, the number of stops per train, departure and end arrivals, and key station stops. A constrained multi-objective high-speed train stop schedule optimization model is therefore established, and an improved genetic algorithm with ideal point method as the fitness function is designed to solve the problem. Finally, a case study of the Beijing-Shanghai high-speed railway is conducted, Results show that the total number of stops after adopting the optimized train stop schedule has been reduced by 7.92%, the train stop cost has been reduced by 19.84%, and the passenger time loss has been reduced by 41.52%. The research results can provide a reasonable reference for the formulation of the train stop schedule.