The southwestern region of China is an ecologically complex area, encompassing fragile geological environments such as plateaus, wetlands, and karst landscapes. Constructing transportation infrastructure in these areas inevitably will cause varying degrees of environmental damage. The extent of the damage and how to restore it are necessary issues for sustainable construction today. However, existing research has rarely conducted intelligent assessments of ecological conditions along railway corridors in southwestern region of China, resulting in limited understanding of how railway network construction affects the ecosystem. Based on machine learning techniques, an intelligent ecological assessment framework for railway corridors was proposed, a long-term ecological evaluation of railway corridors in southwestern region was carried out, and both ecological patterns and the causes of their changes were also analyzed. Results show that from 2001 to 2024, the ecological quality of the southwest railway corridor has shown an evolutionary pattern of overall improvement and local differences. In terms of spatial pattern, Yunnan and Chongqing form high-level ecological core zones, followed by Guizhou and Sichuan, while Xizang lags behind. The construction time of railways does not directly affect ecological recovery time, while complete ecological restoration of railway corridors usually takes one year or even more.

In response to the problems of poor quantitative non-destructive testing of existing rebars and difficulty in synchronously identifying rust and component parameters, a magnetic resonance eddy-current testing (MRECT) imaging system equipped with Mn-Zn ferrite cores was constructed and its resonant high-sensitivity detection mechanism was illustrated via equivalent circuit analysis. With the aid of a three-axis scanning platform and two-dimensional fast Fourier transform (2D-FFT) denoising technology, qualitative grayscale imaging of steel rebars was realized. A segmented voltage corrosion reconstruction method was proposed to quantify the corroded cross-sections, with a relative error of less than 5% for the reconstruction of mild corrosion. By constructing four-point coupled features and adopting light gradient boosting machine (LightGBM), the diameter of rebars and the burial depth of protective layers were simultaneously inverted. By comparing and verifying multi-layer perception (MLP) and random forest (RF), the root mean square error (RMSE) and the mean absolute percentage error (MAPE) of the proposed model's diameter are respectively 0.56 mm and 2.91%. The RMSE of the buried depth is 0.159 cm and the MAPE is 2.36%. A constructed integrated detection scheme of “imaging screening—high-precision corrosion image reconstruction—machine learning quantification” can provide technical support for the non-destructive monitoring of steel rebars in concrete structures.

Ultra-high performance concrete (UHPC) has been widely used in engineering structures owing to its excellent mechanical properties and durability, while its flowability directly affects casting quality and construction adaptability. To reveal the influence mechanism of steel fibers and water-reducing agents on the stacking structure and flowability of UHPC mortar, the variation rules of packing density, water film thickness and relative expansion degree of UHPC mortar with different contents of steel fiber (0%-3%) and dosages of water-reducing agents (1%-3%) were investigated by the wet packing method and expansion degree test. The results indicate that water-reducing agents can release part of the entrapped water by dispersing the flocculated structure among powder particles and increase the packing density and effective water film thickness of UHPC mixtures, thereby improving the flowability of UHPC mortar. However, when the dosage of water-reducing agent exceeds 2%, its role in promoting flocculated water release and flowability improvement gradually tends to become saturated. The addition of steel fibers enhances the spatial barrier effect and wall attachment effect inside the system, destroys the tightly packed structure of particles, and causes some free water to remain on the fiber surface and fiber-particle interface area, resulting in a decrease in effective water film thickness, furthermore, weakening the flowability of UHPC mortar. Based on these findings, a modified water film thickness model considering the dispersing effect of the water-reducing agent was established, enabling effective prediction of the flowability of steel fiber UHPC mortar. The research results can provide a theoretical basis for optimizing the mix proportion of steel fiber UHPC and regulating the construction flow performance.

To meet the demands of complex terrains such as mountainous areas, curved bridges have developed rapidly in recent years. However, the research on the shear connectors of curved corrugated steel web box girder bridges remains limited. To investigate the forcing characteristics of shear connectors of curved corrugated steel web box girder bridges, based on comparative analyses of the overall forcing of straight and curved bridges and the local forcing of shear connectors, two groups of angle steel connector push-out tests characterizing the in-plane direct shear of straight bridge shear connectors and the out-of-plane flexural shear of curved bridge shear connectors, were designed and conducted. The test results show that affected by the out-of-plane effect, the out-of-plane bending moment, out-of-plane shear force, and torque of the curved bridge increase compared with those of the straight bridge, among which, the out-of-plane bending moment is the dominant factor inducing the flexural-torsional coupling effect of the structure. Based on numerical simulations of both the girder segment and the push-out tests, it is obtianed that the complex mechanical behavior of the out-of-plane effect can be equivalently simulated by applying a vertical force on one side of the push-out specimen. In this push-out test, the value of the single-sided vertical load is 0.07 times the direct shear load. The out-of-plane effect will reduce the bearing capacity and ductility of specimens by 4.43% and 28.33%, respectively, and triggers a sharp increase in interfacial slip. For curved box girder bridges with corrugated steel webs, further exploration can be conducted to enhance their crack resistance by appropriately strengthening the inner reinforcement or using high-performance concrete, in order to offset the adverse effects of out-of-plane effect.

To screen out suitable bio-based oils for restoring the properties of aged asphalt, four kinds of bio-oils, soybean oil (SO), waste cooking oil (WCO), palm oil (PO) and tall oil (TO), were selected and the multi-index evaluation and threshold system were studied. Through elemental analysis, gas chromatography-mass spectrometry, Fourier transform infrared spectroscopy, thin-film oven aging, flash point and ignition point, contact angle and adhesion tests, the compositional characteristics, thermo-oxidative stability, construction safety, and interfacial adaptability of different bio-based oils were systematically evaluated. Furthermore, taking the aged asphalt prepared by RTFOT as the object, the effects of different bio-oils on the restoration of penetration, softening point, ductility, and rotational viscosity at 135 °C were analyzed, and theoretical optimal dosage was determined by combining regression analysis. The research results show that all four kinds of bio-oils exhibit good fluidity and construction safety, their flash points all meet the requirements for recycling agents, and the adhesion grade between the recycled asphalt and aggregates all reaches level 5. SO, WCO, and PO are mainly composed of C16～C18 fatty acid components, while TO, in addition to fatty acid components, also has a higher proportion of resin acid components above C18 and carboxylic acid polar functional groups, exhibiting better structural matching and interfacial compatibility. All four kinds of bio-oils improve the penetration and ductility of aged asphalt, reduce the softening point and rotational viscosity. Among them, TO shows the best overall recovery effect on aged asphalt and its theoretical optimal dosage is determined to be 6.5% based on penetration regression. On this basis, a stratified evaluation system for bio-oils consisting of admission threshold, optimization threshold and verification threshold is constructed, providing reference for the screening of bio-based asphalt rejuvenator base oil and the development of composite rejuvenators.

The change in the heat budget caused by highway embankment construction in permafrost regions is the root cause of permafrost degradation and the key control factor in cooling embankment design. To explore the thermal stability characteristics of permafrost embankments, the numerical calculation model based on the typical climatic and geological conditions of the Qinghai-Tibet Plateau was established. Considering four types of annual average ground temperature, four types of embankment width, and four types of embankment height, there are a total of 64 types of combined working conditions. The variation of the heat budget and its influencing factors of permafrost embankment were systematically analyzed. Based on this, BP neural network, GA-BP neural network and GWO-BP neural network were respectively used to establish heat budget prediction models of permafrost embankment. The research results show that the bottom of permafrost embankment exhibits the characteristics of “heat release in cold season and heat absorption in warm season”. Both the heat absorption intensity and duration in the warm season exceed the heat release in the cold season, resulting in a net heat accumulation overall. All three kinds of models produce prediction errors within acceptable ranges, among which the GWO-BP model has the highest prediction accuracy with the lowest mean absolute percentage error. These findings can provide theoretical support and prediction tools for the thermal stability design and active cooling measures optimization of highway embankments in permafrost regions.

Groundwater effect is a key factor for the deformation and instability of anti-dip soft-hard interbedded slopes. However, existing stability analysis methods have not yet systematically incorporated hydrostatic pressure and material softening effects into a unified mechanical framework. The theoretical analysis was conducted on the flexural toppling failure mechanism of anti-dip soft-hard interbedded slopes under groundwater effect. A time-dependent softening coefficient was introduced to characterize the mechanical degradation of soft and hard rocks in water-saturated environments. Based on the cantilever beam limit equilibrium method, a mechanical model of composite cantilever beams capable of distinguishing non-infiltrated, semi-infiltrated and fully infiltrated zones was established, considering the linear distribution of hydrostatic pressure along the rock stratum height. The mechanical equilibrium equations of rock strata under coordinated deformation and free deformation conditions were derived, and a calculation method for the overall safety coefficient of the slope was proposed. The ABAQUS finite element software was used to perform numerical simulation verification on a typical example. Working conditions with and without groundwater effect were compared, and the sensitivity of slope angle, rock stratum dip angle and soft-to-hard rock thickness ratio on the stability was systematically analyzed. The research results show that the slope failure mode is mainly flexural toppling failure, and the error between the safety coefficient obtained by theoretical calculation and the results obtained by numerical simulation is within 10%. Under the given rock stratum and water level conditions, the material softening effect has a more significant influence on stability than hydrostatic pressure. Slopes with a large gradient, a small rock stratum dip angle and a large soft rock thickness ratio are prone to instability.

Taking twin cylindrical piers in a 90° curved channel as the research object, a three dimensional numerical simulation method for water and sediment was adopted to investigate the local scour around twin piers under the action of water-blocking by floating objects. The analysis was focused on the influences of sediment grain size (d), Froude number (Fr), and water-blocking ratio (Sw) on scour-hole morphology, crosssectional scour depth and maximum scour depth around the piers. The results indicate that the sediment grain size has a relatively significant effect on the maximum scour depth around piers under the action of water-blocking by floating objects. When the relative grain size D/d (D is pier diameter) decreases from 4 000 to 133, the maximum scour depth around the piers reduces from 1.27D to 0.41D. Meanwhile, the relative magnitude of the maximum scour depth between the concave and convex banks varies with grain size; under fine gravel conditions, the maximum scour depth at the convex bank is about 18.9% greater than that at the concave bank. An increase in Fr, accompanied by a synchronous rise in the water-blocking ratio, intensifies the coupled action of flow concentration through the bridge opening, flow around floating objects and transverse circulation in bends. Under high Fr conditions, the maximum scour depth at the concave bank pier is 28%~35% larger than that at the convexbank pier. An increase in Sw narrows the effective flow area of the bridge opening, promoting the concentration of water flow towards the bottom of the floating objects and the gap between the two piers. Consequently, the scour-hole morphology has gradually shifted from the relatively independent development of the twin piers to the joint scouring of the bridge-opening area. When Sw=0.40, the maximum scour depth is approximately 2.08 times that under the condition without floating objects. Overall, the water-blocking by floating objects weakens the natural scouring difference caused by water flow in curved channels between concave and convex banks by changing the local flow structure through the bridge opening, serving as an important controlling factor for driving the transformation of the scouring mode of the twin bridge piers. The findings can provide a scientific reference for the protection design and water damage warning of local scour of bridge piers in curved river channels in mountainous areas.

Aiming at the immediate energy replenishment decision-making problem faced by pure electric vehicle users at service-area nodes during long-distance highway travel, existing studies lack sufficient characterization of the phased decision-making behavior of service area nodes for energy replenishment. The influencing variables such as safety margin, information acquisition, user preference and mileage anxiety were introduced to construct a two-stage Logit model for “energy replenishment comparison initiation-energy replenishment scheme selection”. Based on 462 valid stated-preference questionnaire responses, an empirical analysis was conducted. The results show that, in the energy replenishment comparison initiation stage, the estimated coefficient of safety margin is -0.008 9 and is significant at the 1% level, indicating that an increase in safety margin will reduce the probability of users entering the current service area to compare energy replenishment solutions. When 20 km is used as the reference threshold for safety margin segmentation, the coefficients of safety margin in the low-margin and high-margin intervals are -0.012 2 and -0.007 2, respectively, with the absolute value of the former being approximately 1.69 times that of the latter. It is reflected that users are more sensitive to changes in mileage risk under low safety-buffer conditions. In the energy replenishment scheme selection stage, the estimated coefficient of the waiting time is -0.0107, which is significant at the 1% level, indicating that waiting in line constitutes a key service attribute that weakens the attractiveness of the charging scheme in the current service area. The estimated coefficient of safety margin is 0.006 9 and is significant at the 1% level, suggesting that an increase in the safety buffer will improve the acceptability of continuing to go to the next service area. The energy replenishment behavior of highway service areas is not the attribute comparison to a single scheme, but a continuous decision-making process from risk boundary identification to scheme utility balancing.

To address the shortcomings of existing truck-drone collaborative delivery in dynamic demand processing and integrated optimization of customer service time windows, a truck-drone collaborative delivery that could respond to dynamic demands in real-time and meet customer service time window requirements was proposed. Firstly, based on constraints such as truck capacity, drone payload, range, customer time windows and dynamic demand insertion, a dual objective mixed integer programming model that minimized the total logistics operation costs and the number of trucks used was constructed. Secondly, an improved multi-objective particle swarm optimization algorithm was proposed to solve the model, which combined with the scanning algorithm to improve the quality of the initial feasible solution. Subsequently, an external archive update mechanism incorporating simulated annealing acceptance criteria was introduced to enhance the optimization capability of the proposed algorithm. Simultaneously, dynamic demand insertion and resource sharing strategies were designed to achieve a timely response to dynamic customer demands and collaborative shared scheduling of truck-drone routes. Then, the effectiveness of the proposed model and algorithm was verified through comparative analysis with the CPLEX solver, multi-objective adaptive large neighborhood search algorithm, multi-objective genetic algorithm and non-dominated sorting genetic algorithm-Ⅱ. Finally, combined with the real data, an empirical study was conducted to investigate the truck-drone collaborative delivery problem considering dynamic demands and time windows, and the changes in relevant operational indicators under different dynamic demand customer service strategies and different service time period divisions were discussed. The research results indicate that the application of dynamic demand insertion strategy and resource sharing strategy can effectively reduce total logistics operating costs and decrease the number of truck-drone pairs used in the study on truck-drone collaborative delivery considering dynamic demands and time windows. The research results are helpful for the construction of urban truck-drone collaborative logistics distribution system and can provide technical and methodological support for the multi-level truck-drone logistics distribution network scheduling problem considering diverse dynamic customer demands.

To investigate the impact mechanism of cognitive distraction on driving safety in potential vehicle-pedestrian conflict scenarios, two typical hazardous scenarios were selected, that is, potential illegal crossing pedestrians (scenario I) and blind areas at unsignalized crosswalks (scenario II). Using a driving simulator, eye-tracker and physiological instrument, the eye-movement, electrocardiogram and vehicle control data of drivers were collected under three driving states such as normal driving, 1-back task and continuous operation task. The influence of cognitive distraction on the safety of driving behavior was systematically analyzed. The driving safety measurement indicators were selected, and the CRITIC-entropy weight method and grey relational model were combined to conduct comprehensive quantitative evaluation of driving safety level. The robustness of the model was verified. Research results show that cognitive distraction significantly delays drivers’ visual perception of hazards, increases the gaze response time ratio to potential danger areas, raises drivers’ psychological load, delays braking response time and decreases operational stability. In scenario I, the safety degree of the 1-back and continuous operation state is reduced by 12.33% and 18.48% respectively, compared to that of the baseline state; while reduced by 10.28% and 14.88% respectively in scenario II. The continuous operation task has a greater impact on driving safety under the conditions of this experiment. These findings provide a theoretical basis for formulating driving regulations and designing targeted safety training programs.

Aiming at the problem that real-world regional traffic flow data exhibits pronounced non-stationarity and complex spatiotemporal dependencies, which causes existing prediction models to suffer from degraded generalization ability, a regional traffic flow prediction model integrating statistical residual enhancement and dynamic graph spatiotemporal modeling (AMSR-STNet) was proposed. By fusing residual statistical enhancement (RSE) and reversible instance normalization (RevIN) to construct historical statistical priors and combining an adaptive dynamic graph convolution with a Top-K sparsification strategy alongside a multi-scale dilated temporal convolutional network (MSD-TCN), the proposed model was enabled to dynamically capture latent topological correlations among nodes and extract long-term and short-term temporal dependencies. Thereby, the proposed model’s capability to suppress data distribution shifts and model local high-frequency temporal fluctuations was enhanced. Regional-level short-term traffic flow prediction was conducted by taking four public datasets representing various scales of road networks as an example, namely PeMS03, PeMS04, PeMS07 and PeMS08. The research shows that compared to traditional graph convolutional networks such as DCRNN and AGCRN, as well as baseline models based on attention mechanisms and the latest state space (Mamba), the AMSR-STNet model exhibits competitive predictive performance in metrics including mean absolute error (EMA), root mean square error (ERMS), and mean absolute percentage error (EMS), and achieves optimal results in some metrics. Furthermore, long-horizon multi-step prediction scenarios and typical non-stationary fluctuating nodes are selected to conduct verification and ANOVA tests, and the AMSR-STNet model maintains both smaller prediction errors and more acute tracking capability for high-frequency sudden changes. It is demonstrated that the generalization and anti-interference ability of the proposed model under different scales of road networks and complex non-stationary situations can provide methodological references for predicting regional traffic conditions.

To address the problems of insufficient recognition accuracy and unclear boundary representation of drivable areas for underground unmanned mining vehicles under low illumination, dust interference, and weakened roadway boundaries, a drivable area perception method based on image self-enhancement semantic feature segmentation and attention mechanism fusion was proposed. Firstly, a GACL image enhancement method was constructed by integrating HSV brightness separation, adaptive Gamma correction, and contrast limited adaptive histogram equalization (CLAHE). Secondly, a semantic feature fusion (SFF) module was introduced to enhance the interactive expression between shallow boundary details and deep semantic information, and a convolutional block attention module (CBAM) was embedded to recalibrate channel and key spatial features. With SegFormer-B0 as the backbone, a semantic segmentation network enhanced by semantic feature fusion and convolutional block attention (SFFC-SegFormer) was constructed. Finally, the main connected component was extracted from the segmentation mask, and a pixel-level road centerline was generated by interval sampling of candidate center points and distance-constrained robust fitting. Its representation was then transformed into a physical reference path based on local road-plane homography transformation. The research results show that the proposed model achieves a mean intersection over union of 94.28% and a mean F1-score of 95.82% on a self-built underground roadway dataset, with a parameter count of 7.54 M and an inference speed of 28.4 FPS, which maintains high segmentation accuracy while having a small model parameter size. The proposed method can improve the segmentation integrity of drivable areas in complex underground degraded environments and provide an effective visual perception foundation for local reference path extraction.

To address the issues of freezing of preceding-vehicle state information for cooperative adaptive cruise control (CACC) vehicles after communication interruption and inaccuracy of preceding-vehicle information in model predictive control (MPC), a CACC-MPC control method based on temporal convolutional network (TCN) and velocity residual prediction of the preceding-vehicle was proposed. First, a longitudinal discrete prediction model of CACC platoon, consisting of one leading vehicle and three following vehicles, was established. And communication link state, packet reception state and information age were introduced to describe the degradation process of preceding-vehicle information after communication interruption. Then, a gated baseline speed was constructed according to communication credibility, and a historical feature sequence was formed by use of the status of the ego vehicle, radar observations, communication maintenance status and communication degradation information. TCN was used to predict the speed residual and combined with the gated baseline speed to generate the preceding-vehicle speed preview sequence required by MPC. Finally, based on Simulink/CarSim co-simulation platform, verification was conducted under conditions such as normal communication, communication delay, superposition of communication delay and packet loss, as well as communication interruption of different durations. The results show that after communication interruption, the communication retain speed cannot reflect the subsequent movement changes of the navigation vehicle, and the proposed method can reduce the influence of frozen communication information on preceding-vehicle speed preview, maintain a basic safe spacing from the followed vehicles, and reduce the control input and its fluctuation of change rate, improving the speed following, vehicle distance maintenance and control smoothness of the vehicle queue under communication degradation conditions.

To enable personalized and safe lane-changing decisions for autonomous vehicles in multi-vehicle interactive environments, a game theory-based multi-cost lane-changing decision model was proposed. The total decision cost comprised three sub-costs, namely safety, comfort and traffic efficiency, and a driving style recognition mechanism was introduced to support personalized lane-changing games among multiple vehicles. Principal component analysis was employed to extract features from drivers’ lane-changing data, K-means clustering was used to classify driving styles, and the convolutional neural network was adopted for style recognition. The interaction and decision-making process of lane changing between vehicles was modeled as a strategic game, through which the lane-changing strategy minimizing the total decision cost was derived via game-theoretic analysis. Typical highway scenarios were selected to conduct two-stage game simulations for four driving style combinations, to accomplish overtaking maneuvers. The research results indicate that when the driving styles of the ego vehicle and the obstacle vehicle tend to be consistent, the “cooperative” effect between the two parties in the game is enhanced. Specifically, compared with combination 2 #, combination 1 # reduces the total cost of the first stage game by 8.80% and that of the second stage game by 4.31%, respectively; compared with combination 4 #, combination 3 # reduces the total cost of the first stage game by 7.01% and that of the second stage game by 2.57%, respectively. When the ego vehicle’s speed is substantially higher than that of the obstacle vehicle, the game cost decreases significantly, with the second-stage game cost respectively reduced by 50.50%, 52.80%, 41.33% and 44.00% for the four combinations, compared to the first-stage game cost. At this time, the obstacle vehicle is more likely to accept lane changing, significantly reducing conflicts and risks during the lane changing process, and improving traffic efficiency. The proposed game-theoretic lane-changing decision model forms a targeted technical pathway for driving style recognition, multi-style strategic game modeling, and personalized trajectory planning, which can support optimal lane-changing decisions for autonomous vehicles in complex traffic environments.