[Objective] In the context of global warming, the existing highways in permafrost regions of Qinghai-Tibet Plateau are facing a series of challenges, i.e., surface settlement, roadside water accumulation, groundwater saturation, melting interlayer thickening, and permafrost ongoing degradation. To address the severe challenges to in-service subgrade defects treatment and modification,it is urgent to propose more refined and adaptable design and disposal concepts beyond current design principles. [Method] The development of foundation design methods for permafrost regions abroad and the evolution of highway subgrade design methods for permafrost regions in China were reviewed. Based on latest field investigation data, the causes and contents, should be improved in current design system of subgrade settlement on existing highways in permafrost regions of Qinghai-Tibet Plateau, were analyzed. [Result] It is recommended to develop and enhance the third design principle, named proactively improving foundation conditions, based on the two existing principles, i.e., protecting permafrost and allowing permafrost thawing. The concept of treating permafrost foundation is introduced, along with a corresponding collaborative design method for pavement-subgrade-foundation system in permafrost regions. The influences of foundation property changes (induced by climate change) on subgrade deformation and stability are finely investigated. [Conclusion] According to the categorization, characteristics and engineering applications of treatment techniques for current subgrades, 0-3 m shallow foundations, and deep foundations over 3 m, the urgent technical challenges are highlighted, i.e., the refined hydrological and geological survey, the improvement on engineering geological evaluation index system, the establishment of long-term efficacy characterization system for permafrost foundation treatment methods, and the development of new treatment materials and equipment.

[Objective] This study investigated the lane-changing intention and trajectory prediction of autonomous vehicles in mixed traffic flow scenarios in expressway interweaving zone. A novel Gra-Informer model with dynamic spatiotemporal fusion architecture was proposed. [Method] First, the model extracted high-dimensional intention features by using a graph neural network based on vehicle interaction topology. The intention-aware based spatiotemporal representations were established. Then, the encoder employed probabilistic sparse attention, combined with temporal distillation, to achieve the efficient long-sequence feature extraction and compression. Finally, the decoder used intention features as conditional priors to generate the vehicle trajectory predictions. [Result] The model was evaluated by using Savitzky-Golay filtered natural driving data, demonstrating 91.54% accuracy in lane-changing intention prediction. In the performance comparison, the model shows the prominent spatiotemporal features for modeling ability. During 3 seconds short-term prediction, the average displacement error reduced by 12.32%-30.06%; and the terminal error and the misjudgment rate both reduced by 4.00%-30.29%. During 5 seconds long-term prediction, the average displacement error reduced by 17.51%-33.33%; the terminal error reduced by 17.19%-33.47%; and the misjudgment rate reduced by 17.50%-33.50%. These outcomes demonstrate the advantage of modeling temporal coupling characteristics in interactive scenarios. [Conclusion] Gra-Informer model can effectively capture the dynamic interaction characteristics between autonomous vehicles and traffic vehicles through the spatiotemporal joint modeling and the intention-trajectory coordination optimization. The model introduces the prior intention information as conditional constraints for the trajectory prediction. It not only maintains the physical feasibility, but also significantly improves both short-term and long-term prediction accuracy in complex traffic scenarios.

[Objective] The study investigated to accurately simulate the traffic flow characteristics in expressway curve sections in mountainous regions, and alleviate the traffic problems at curves by using autonomous driving technology. A two-lane cellular automaton model, considering the mixed traffic of manually driven and autonomous vehicles in curve sections, was established. [Method] The correctness of model was verified through curve radius and road friction coefficient, considering the influence of different curve conditions on traffic flow. The influence of curve section transition curves was proposed. The influence degree of different proportions of transition curves on traffic flow in curve sections was analyzed to improve the traffic flow simulation environment. On this basis, the scenarios with mixed connected and automated vehicles (CAVs) were introduced to explore the influence of different CAV penetration rates on traffic flow in curve sections. [Result] Different proportions of transition curves have the significant influence on traffic flow in curve sections. The traffic efficiency at curves is optimum when the ratio of three elements (i.e., curve section transition curve, circular curve, transition curve) is close to 1∶1∶1. The traffic congestion at curves is gradually alleviated when CAVs appearance on roads. When all vehicles are CAVs, the number of lane-changing drops to 0, and no congestion occurs anymore. In addition, a unique traffic flow plateau phenomenon is observed in the traffic flow in curve sections, i.e., the vehicles run stably for a long time in certain density conditions, and the traffic flow remains a fixed value without varying with density. [Conclusion] The proposed model effectively simulates the traffic flow characteristics in expressway curve sections in mountainous regions, and clarifies the optimal proportion of transition curves and the positive role of CAVs.

With the continuous advancement of sensing, communication and computing technologies,and coupled with rapid deployment of intelligent transportation infrastructure, the cooperative autonomous driving is becoming a key development direction in intelligent connected vehicle (ICV) systems. This paper aims to systematically review the core technical architecture underpinning the integration of smart highways and cooperative autonomous driving, while summarizing the current progress and emerging trends in this field. First, it introduces the conceptual evolution of smart highways, their classification framework, and their functional role in supporting autonomous driving, along with representative development paths and engineering practices in major countries and regions. Second, it analyzes the development status and limitations of core single-vehicle autonomous driving technologies, highlighting their performance bottlenecks in complex scenarios. Third, the paper focuses on key enabling technologies for cooperative autonomous driving,e.g., vehicle-road-cloud integration, cooperative communication, sensor fusion, and joint decision-making and control;and discusses the challenges of multi-source data fusion and mixed traffic flow management. Finally, it reviews current testing and validation approaches for cooperative driving systems, and addresses the major issues in their engineering implementation, followed by an analysis on future development trends in smart highway-vehicle cooperative systems.

[Objective] The study investigated the influence of heavy-duty vehicles flow on the on-ramp system. The heterogeneous traffic flow model was developed for the on-ramp, aiming to examine how heterogeneous heavy-duty vehicles affect the traffic flow characteristics in merging areas. [Method] An improved two-lane cellular automaton model was employed in this study. By analyzing the characteristics of vehicle driving behaviors, the lane-changing rules were systematically optimized. The heterogeneous traffic flow model for the on-ramp system was developed with the aim of investigating how the traffic flows on main road and ramp affect the traffic characteristics in merging areas in heavy-duty truck operating conditions. [Result] As the proportion of heavy-duty trucks increases, the traffic congestion in merging areas of on-ramp system becomes more pronounced, accompanied by a corresponding reduction in the saturation flow rate. When the on-ramp entry probability and the proportion of heavy-duty trucks remain unchanged, the traffic flow in merging areas increases with the main road entry probability, which is below 0.4. The increase of heavy-duty trucks proportion on the main road negatively influences the performance of merging areas. Both the critical entry probability from main road and the saturation flow exhibit the decreasing trend as the proportion of heavy-duty trucks on main road increases. When the main road entry probability and the proportion of heavy-duty trucks remain unchanged, the speed in merging areas decreases as the on-ramp entry probability increases, which is below 0.3. The increase of heavy-duty trucks proportion on the ramp has a relatively small influence on the flow in merging areas, but has the significant effect on the speed. [Conclusion] The model, considering on-ramp heterogeneous traffic flow, can effectively reduce the probability of on-ramp traffic congestion.

[Objective] The study investigated the influences of high complexity of road network traffic spatio-temporal dependencies on traffic flow prediction accuracy, as well as the similarities across different time scales and the dynamic correlation among road network nodes over time. The traffic flow prediction model was proposed based on the multi-scale downsampling convolutional interaction dynamic graph convolution neural network. [Method] In the spatial dimension, the dynamic spatial correlation was captured with dynamic graph diffusion convolution modules in the traffic network. The modules generated an adjacency matrix from the input traffic flow data and a predefined graph structure. This matrix fused with an adaptive adjacency matrix to form a dynamic adjacency matrix, so as to capture spatio-temporal features of network nodes. In the temporal dimension, taking the data as inputs, which were from adjacent time periods, same period of previous day, and same period of previous week. These data reflected the temporal similarity of traffic flow data across multiple time scales. The input sequences with three time scales were divided by intervals based on time series proximity and particularity. The temporal correlation of traffic flow data was captured by adopting an interactive learning structure. During this process, the dynamic graph diffusion convolution modules were embedded into the interactive learning structure to synchronously capture spatio-temporal correlations. Furthermore, the inherent causal relation between transport network dynamic variation and time was investigated. To address non-strict periodicity in traffic flow data, an attention mechanism was used to resolve periodic offset issues. Finally, to validate the effectiveness of the proposed model,a comparative analysis was conducted with selected classical and state-of-the-art traffic flow prediction models.[Result] Compared with some deep learning baseline models, the proposed model prediction result indicates the performance improvement of 4.5%-33.7% in mean absolute error of regression prediction accuracy indicator. [Conclusion] The study result will effectively enhance the traffic flow prediction accuracy, and provide novel approach for dynamic traffic management in ITS.

[Objective] To address the initial support failure caused by large deformation during tunnel excavation through weak rock strata in rapid topographic change terrains, this study introduced the combined resistance-release concept of resistance-limiting and energy-dissipating support. Taking a tunnel on Zhenxiong-Hezhang expressway as engineering case, the study investigated the structural design and application of resistance-limiting and energy-dissipating support. [Method] A numerical model of rapid topographic change terrain was established to analyze the causes of large deformations. The steel-plate resistance-limiting and energy-dissipating support was simplified into the I-shaped structure. Numerical simulations and model tests were conducted to analyze the force-deformation characteristics of I-shaped steel resistance-limiting dampers. The key parameters were determined, e.g., peak resistance, constant resistance, and compressibility ratio. An initial support scheme was designed and validated through field applications. [Result] The primary cause of large deformation is identified as the downward transfer of mountain self-weight stress in rapid topographic change terrains. The I-shaped steel resistance-limiting dampers exhibit the staged compressive deformation, and provided the bearing capacity exceeding 0.91 MPa. The field tests demonstrate that after stress release via damper deformation, the maximum stresses on initial support steel frame and shotcrete are 23.1 MPa and 20.3 MPa respectively, both below the material strength limits. [Conclusion] The initial support structure integrated with I-shaped steel resistance-limiting dampers effectively releases surrounding rock stress through controlled deformation. It reduces the internal stress in support system, and ensures construction safety and quality.

[Objective] To improve the early warning capability for rockfall disasters on highway slopes and ensure road traffic safety, a risk assessment model for rockfall hazards was constructed. [Method] The study selected multiple analytical evaluation indicators from four aspects( i.e., unstable rock conditions, slope conditions, surface conditions, and other external factors, covering key geological, geomorphological, and environmental elements. The analytic hierarchy process), combined with the entropy-based disaster model, was employed to address the complex relations among these indicators. It integrated both expert subjective judgment and data objectivity for improving the scientific rigor and accuracy of the assessment system. Furthermore, referencing national standards and empirical data, the hazard classification standard for rockfall disasters was established. The risks were categorized into four levels, i.e., red (extremely high risk), orange (high risk), yellow (moderate risk), and blue (low risk). [Result] The model exhibits high accuracy in assessing rockfall hazards. The regions classified as red and orange closely align with the historically high-frequency rockfall disaster regions, while yellow and blue regions correspond to the regions with lower or negligible disaster occurrence. The systematic quantitative analysis on rockfall hazards along highways was achieved by using this model, providing a scientific basis for slope protection design and disaster early warning. [Conclusion] The rockfall hazards can be effectively evaluated with the proposed method. It offers the robust support for future monitoring and early warning systems for rockfall disasters on highway slopes, significantly contributing to the improvement of safety protection measures along roadways.

[Objective] The study investigated various construction schemes to improve the safety and economy of tunnel entrance construction for large-cross-section small-interval tunnels with different spans. The study investigated the reasonable tunnel entrance construction scheme for large-cross-section small-interval tunnels with different spans. [Method] Shangxinguang tunnel project for Urumqi-Yuli expressway in Xinjiang was taken as the study case. Three-dimensional mechanical model of entrance section of small-interval tunnels with same span was established with finite element software MIDAS GTS NX. The surrounding rock stability was studied with three construction schemes, i.e., center diaphragm method, double side drift method, and three-bench method. [Result] The surface settlement variation pattern with different construction schemes is basically consistent. The maximum values of surface settlement in both left and right tunnels appear at the right arch shoulders. The three-bench method has the worst control effect on surface settlement. As the excavation depth increases, the variation trend of right tunnel vault crown settlement with different schemes are roughly the same. The final settlements are ranked from large to small as three-bench method (23.12 mm), double side drift method (19.96 mm), and center diaphragm method (17.49 mm). While the stress release rate of surrounding rock is the fastest when using three-bench method for construction. The surrounding rock maximum plastic strains with three excavation methods are ranked from large to small as three-bench method (2.35×10^-2), center diaphragm method (6.29×10^-3), and double side drift method (5.61×10^-3). The plastic zone penetrates when using three-bench method. [Conclusion] By using both center diaphragm method and double side drift method can effectively control the surrounding rock stability, however the construction is slow with double side drift method. Taking into account factors (e.g., surrounding rock stability, construction cost, safety, and construction period), the center diaphragm method is ultimately selected for construction. The in-situ monitoring shows that the tunnel vault crown daily average settlement is 0.51 mm, which meets the requirements. It can provide references for similar projects.

[Objective] The study systematically investigated the influences of different activation methods on the structure of desulphurized rubber powder and the properties of modified asphalt. Three activation processes (i.e., microwave, chemical, and twin-screw extrusion) were applied to pre-treat the waste rubber powder, and prepare the corresponding modified asphalts. [Method] The chemical composition and microstructure of activated rubber powders were characterized by using Fourier transform infrared spectroscopy, scanning electron microscopy, and thermogravimetric analysis. Simultaneously, the dynamic shear rheometer and bending beam rheometer tests were performed to evaluate the high-and low-temperature performances of modified asphalts. [Result] While the microwave and chemical activations effectively break the unsaturatedCCbonds in the rubber powder, they have limited effects on the destruction of crosslinked network structure. In contrast, the twin-screw extrusion activation significantly disrupts the polysulfide crosslinks through shear forces, enhancing the solubility of rubber powder. The rheological tests reveal that the rubber powder with high solubility improves the low-temperature cracking resistance of asphalt, but the main chain fracture leads to 30% decrease in high-temperature rutting resistance. Moreover, the twin-screw extrusion activation promotes the refinement of rubber powder particles and the generation of surface functional groups, but the excessive desulphurization weakens its mechanical properties. [Conclusion] Different activation processes affect rubber powder structure and modified asphalt performance through distinct pathways. The microwave and chemical activation are suitable for scenarios requiring high-temperature performance, while the twin-screw extrusion activation has significant advantages in improving low-temperature crack resistance. These findings provide the theoretical guidance for selecting activation processes and precisely controlling the performance of rubber powder modified asphalt.

[Objective] Comprehensively understand the research progress of metrological testing technology in the field of highway transport. [Method] First, from the perspective of metrological testing for essential elements of highway transport, the specific keywords were searched through core database of Web of Science. A total of 15 264 papers from 2014 to 2023 were retrieved, for which the year-by-year distribution characteristics were analyzed. Second,bibliometric method and scientific knowledge map were used to visualize the papers in past 5 years. The research hotspots were summarized by keyword co-occurrence and correlation strength clustering. Finally, the future research direction was put forward on the basis of research status and technical development trend in the field of detection, monitoring and metrology, e.g., infrastructure, traffic equipment, traffic participants and traffic environment. [Result] The research of metrological testing technology in the field of highway transport has developed rapidly in recent years. Current research hotspots include pavement damage detection or monitoring, bridge and tunnel structure monitoring, vehicle motion characteristics detection, driving situation awareness, traffic noise monitoring, driver fatigue state detection, etc., especially in the cross-fused research of machine vision and artificial intelligence, which has broad prospect. [Conclusion] In the future, these research fields will remain hot, i.e., infrastructure inspection or monitoring and damage prediction, metrological testing and performance evaluation on intelligent sensing system, traffic participant behavior analysis and intention prediction, highway traffic carbon monitoring and measurement, and quantitative evaluation on vehicular traffic situation awareness system.

[Objective] Accurate traffic volume prediction is the basis for achieving expressway active control. However existing methods suffer from insufficient prediction accuracy and stability due to neglecting spatio-temporal correlations in traffic volume and external factors, e.g., weather conditions. To fully explore the intrinsic connection between traffic volume and external features as well as the complex spatial-temporal correlation of traffic volume itself, a novel feature-fused spatio-temporal graph convolutional networks (FSTGCN) was proposed for traffic volume prediction. [Method] The feature convolutional networks, integrating traffic volume and external features, were established with FSTGCN. It was used to capture correlations between traffic volume and external features in feature modules. The graph convolutional networks based spatial modules were employed to extract spatial correlation of traffic volume. Finally, the outputs from feature module and spatial module were connected and input into Gated Recurrent Unit based temporal modules for spatio-temporal correlation learning. [Result] Experimental results with real data from expressways in Shandong province indicate that FSTGCN significantly outperforms mainstream benchmarks in long-term prediction tasks, while maintaining superior stability across both short-and long-term horizons. Ablation tests confirm the positive contributions of feature module, spatial module and temporal module. The extreme error comparison shows that FSTGCN model has strong robustness. Simultaneously, the training efficiency analysis further validates its practicality for real deployment. [Conclusion] The FSTGCN effectively addresses accuracy and stability challenges in expressway traffic volume prediction through multi-source feature fusion and spatio-temporal correlation modeling.

[Objective] To accurately predict the unconfined compressive strength of cement stabilized recycled aggregate, so as to shorten the mix design period, the LSTM-based cement stabilized recycled aggregate strength prediction model was established. [Method] The particle swarm optimization (PSO) was used to perform a global search for the best parameters of LSTM model. The test data of unconfined compressive strength of cement stabilized recycled aggregate with different ages and mix proportions were used as the dataset. Eleven variables, e.g., cement content, moisture content, recycled aggregate replacement rate, and curing age, were used as the model input layers. The unconfined compressive strength was used as the output layers. The evaluation indicators were introduced to analyze the performance of established LSTM model and PSO-LSTM model. The eleven input layers were divided into material influences and non-material influences. The correlation coefficients were introduced to study the correlation between variables and unconfined compressive strength. [Result] The accuracy of LSTM model and PSO-LSTM model both reaches over 98%, and both of the models can accurately predict the unconfined compressive strength of cement stabilized recycled aggregate. MSE, RMSE, MAE, MAPE, R² of models indicate that PSO-LSTM model has higher accuracy, smaller error rate and better fitting effect. According to the model prediction results, the strength of cement stabilized recycled aggregate with different proportions was compared. The most suitable cement content, recycled aggregate replacement rate and skeleton type were obtained. [Conclusion] By introducing PSO, the accuracy and precision of LSTM model can be effectively improved for strength prediction on cement stabilized recycled aggregate.

[Objective] Traditional expressway service areas have limited functions. Their integration with regional economies is lack of synergy. Transforming them from closed operations to opened-up development is widely recognized. In this context, the scientific evaluation on opened-up development potential of expressway service areas has become the key premise for the service areas development according to local conditions. [Method] First, the typical cases and literature on opened-up service areas were analyzed. The evaluation indicator system was established from four aspects, i.e., locational conditions, socio-economy, tourism resources, and project conditions. Next, the entropy weight method was used to determine the indicator weights. TOPSIS method was applied to calculate the comprehensive scores for opened-up development potential. K-means clustering algorithm was adopted to classify the results. Finally, 84 service areas located at 14 expressway sections in Hebei Province were selected. These service areas were chosen for high traffic, modern facilities, and abundant resources. Their opened-up development potentials were thoroughly evaluated. [Result] Based on the distributed locations and characteristics, 15 pilot service areas have been selected out. According to the principle of One Area, One Feature, various unique development themes have been defined for different service areas. [Conclusion] The service areas’ opened-up development relies not only on transportation infrastructure, but also closely links to the regional economies, tourism vitality, and cultural resources. 15 service areas with great development potential, distinctive features, and balanced overall distribution in Hebei Province can be prioritized as the pilot service areas

[Objective] To solve the problems of easy agglomeration and poor dispersion of graphene oxide (GO) in cement, the comprehensive performance of composite material was improved by using modifiers. [Method] Silane coupling agent KH570 was used as raw material to modify GO, and the modified GO (KGO) was prepared. The dispersion of GO and KGO was studied, and the influences on hydration process and mechanical properties of cement were studied. Meanwhile, the morphological characterization of samples was analyzed by using X-ray diffraction, scanning electron microscope, Fourier transform infrared spectroscopy, and Ultraviolet-visible spectroscopy. Additionally, the comprehensive thermal analysis was performed by using differential scanning calorimetry and cement hydration calorimetry. [Result] The optimum modification temperature of KH570 for GO is 50 ℃, and the resultant modified product is KGO50. It has great dispersion in simulated pore solution. Compared with GO cement composite, the improvement amplitudes of 28-day flexural and compressive strengths of KGO50 group are more significant. The hydroxyl and epoxy groups on GO surface are consumed during modification process while introducing silane groups, thereby GO can uniformly disperse in cement paste under the action of electrostatic repulsion. KGO50 retains the toughening effect of GO, promotes the hydration of cement, increases the content of Ca(OH)₂, and effectively regulates the hydration products. [Conclusion] KGO50 modification technology combines both economic and performance advantages. It has certain engineering significance and canprovide theoretical basis for the development of high-performance cement.

[Objective] Modular buses provide novel approaches to promoting the sustainable development of public transportation. This study aims to introduce the modular buses, i.e., a new form of public transport. It uses the advantages of flexible module coupling and decoupling, as well as the seamless transfer, to eliminate the transfer waiting time and optimize the bus route design. [Method] First, the main operational issues of conventional bus systems were examined. The study compared traditional bus in-station transfer with modular bus seamless transfer. It summarized the operational advantages of modular buses, and identified the key considerations for path design. Second, a line design model was developed with modular buses as the study object. With the dual objective of reducing operating costs and passengers’ generalized travel costs, a modular bus operation route design model considering passenger path assignment was constructed. Third, in the conditions of established set of running routes, set of path candidates and travel demand, the model was applied to obtain the modular bus running design and the corresponding passenger path allocation scheme. Finally, the example analysis was carried out by adopting a simple traffic network; and the sensitivity analysis on the model was carried out by changing the line length limit and adjusting the weight of each part of the objective function of model. [Result] The proposed model produces reasonable line designs and passenger path assignments. The computation time is within seconds. The passenger transfers are kept within two times per trip. [Conclusion] The operating lines design and passenger paths assignment obtained with the proposed model are helpful for modular bus to reduce the operation cost and generalized travel cost at the same time, which has the practical application values.

[Objective] This study proposed a multi-modal data fusion method for remaining useful life (RUL) prediction on expressway electromechanical equipment. The goal was to improve the equipment life prediction accuracy and robustness, and to provide the reliable support for proactive maintenance. [Method] First, a dataset named SW-RUL-DATAS was built, covering the transient voltage and current, temperature and humidity, maintenance records, and system logs. Second, the features were extracted separately from different modal data, including environmental data extraction with convolution, electrical signals extraction with frequency and time-domain analysis, and text data extraction with BERT. Third, an autoencoder was applied for feature fusion and dimensionality reduction,yielding the unified,high-quality feature representation. Finally, the fused sequence features were fed into the recurrent neural network. The hyperparameters were optimized by using AutoML to achieve the end-to-end RUL prediction. [Result] The tests on three sub-datasets indicate that the proposed method outperformed Kaplan-Meier, ARIMA, CNN+GRU and CNN-LSTM in both MSE and RMSE. The model can accurately predict potential faults of cameras, lighting systems, and other equipment 2-5 days in advance, with accuracy and recall improved by 5%-15% compared with baselines models. Moreover, the data sensitivity tests show that the transient electrical signals, maintenance logs, and environmental data all make significant contributions. The multi-modal fusion is the key factor for performance improvement. [Conclusion] The proposed multi-modal fusion prediction framework effectively integrates heterogeneous multi-modal data. It achieves higher accuracy and robustness in expressway electromechanical equipment RUL prediction. The study highlights the importance of transient electrical features and maintenance logs, as well as shows the effectiveness of deep learning with AutoML. The method provides the reliable data and technical support for the predictive and proactive maintenance of expressway electromechanical equipment.

[Objective] The study investigated to reduce energy losses caused by tire-pavement rolling resistance on asphalt pavement while meeting the performance requirements for road construction. The response surface methodology was used to incorporate PE/PP elastic modifiers, SBS modifiers, and rubber oil into base asphalt to prepare PPSE (polyethylene polypropylene SBS elastic) modified asphalt. [Method] The technical performance of PPSE modified asphalt was evaluated through high-temperature gel permeation chromatography tests, low-temperature bending beam rheometer tests, and dynamic shear rheometer tests. The road performance advantages of PPSE modified asphalt mixtures were investigated through rutting tests, low-temperature beam bending failure tests, pendulum-type skid resistance tests, and dynamic modulus tests. The tire-pavement contact tests were carried out by using pressure-sensitive film technology. The rolling resistance was evaluated in conjunction with contact mechanics theory. [Result] PPSE modified asphalt exhibits excellent technical performance, effectively functioning as a binder. PPSE modified asphalt mixtures demonstrate outstanding high-temperature stability and low-temperature crack resistance; the dynamic modulus performance contributes to reducing rolling resistance; and the skid resistance meets the road acceptance requirements. The contact stress distribution between PPSE modified asphalt pavement and tire is significantly non-uniform. The high shear modulus of PPSE modified asphalt within temperature-time domain enables stronger cohesion among aggregate particles under the same static load. That will form more stable mixture skeleton force chain, reduce adhesion by 21.12%, and significantly lower the rolling resistance (viscous force). [Conclusion] PPSE modified asphalt has excellent viscoelastic mechanics and cohesive characteristics, which improves the deformation coordination uniformity of its mixture, modifies the tire-pavement contact stress distribution, and reduces the tire rolling resistance.

[Objective] Improve the safety of construction and completion for box girder bridge with corrugated steel webs. [Method] Aiming at web force difference of single-box double-cell box girder with corrugated steel webs, the 3D finite element analysis and birth-death element method based on actual bridge model were adopted. The longitudinal bending moment distribution, section shear force distribution and web longitudinal stress distribution during construction and completion process were studied. Simultaneously, the comparative analysis was conducted on force distribution and shear lag effect of box girders with similar cross-sections. Finally, the comparative verification was conducted with actual bridge experiment. [Result] Referring to the web-based partition method commonly used in grillage method, the proportions of high, middle and low webs in longitudinal bending moment lateral distribution are approximately 30%, 42% and 28% respectively. In shear force distributions, the proportion of middle web is the highest, the proportion of low web is the lowest, and the difference between high web and low web is about 2%. The maximum shear lag coefficients at top plates of concrete box girder and box girder with corrugated steel webs are 0.858 and 0.896 respectively; and the coefficients are 0.95 and 0.97 at base plates. The closer to the steel-concrete section, the greater the longitudinal stress increment of middle web. Away from the steel-concrete section, the stress increment of high web is the largest, followed by the middle web, and the low web is the smallest. [Conclusion] The longitudinal bending moment lateral distribution and section shear force indicate that the middle webs account for the largest proportion, followed by the high webs, and the low webs account for the smallest. Compared with box girders with same kind of section, the web difference in box girder bridge with corrugated steel webs has less effect on longitudinal bending moment lateral distribution, and more effect on web plate shear distribution. Using corrugated steel web instead of concrete web can reduce shear lag effect of box girder. During dynamic construction process, the longitudinal stress increment of nearby middle web is larger than that of nearby side webs due to the influence of steel-concrete composite segment.

[Objective] The water log at expressway superelevation transition section is a common hidden danger to traffic safety. The catchment path and water log distribution characteristics at superelevation transition sections are essential for improving road safety. [Method] A method was developed to calculate the maximum catchment path length based on the principle of maximum resultant gradient. The surface rainfall was simulated by using discrete phase model and Eulerian wall film model. The influences of various geometric parameters on catchment path and water log distribution characteristics at superelevation transition sections were systematically investigated. [Result] The maximum catchment path generally follows a parabolic shape. Based on different starting points of path, the flow directions are classified into two patterns, i.e., median strip-hard shoulder edge-median strip, and hard shoulder edge-median strip-hard shoulder edge. The maximum catchment path length increases with the longitudinal gradient. When originating from the median strip, the maximum catchment path length growth is fast with smaller longitudinal gradient. The maximum catchment path length increases as the superelevation runoff decreases, especially with steeper longitudinal gradients. The ratio of catchment path lengths across different road widths corresponds to their geometric proportions. Water log distributions along maximum catchment path show three patterns with longitudinal gradient variation, i.e., single peak, double peak, and continuous increase. The superelevation runoff and road width mainly affect the absolute thickness of water film, but have limited influence on the water log distribution trend. [Conclusion] This study quantitatively reveals the influence of geometric parameters on catchment path and water log distribution characteristics at superelevation transition sections. The results provide data support for the catchment design of superelevation transition sections.

[Objective] To provide references for the composite reinforcement design and construction of RC beams, the prestressed carbon fiber-reinforced polymer (CFRP) bars embedded RC beams with textile reinforced concrete (TRC) layer were adopted. This study investigated the influences of prestress state and different reinforcement methods on the bending resistance of beams. [Method] The four-point bending tests were conducted on one conventional RC beam and three reinforced beams. The failure modes, load levels at different stages, crack distribution and width, as well as the strain of reinforcing materials of the reinforced beams were analyzed.A sectional analysis method was employed to develop the bearing capacity formula for RC beams reinforced with TRC and embedded CFRP bars. [Result] Compared with the unreinforced RC beams,the composite reinforced beams exhibit significant higher cracking, yield,and ultimate loads. The magnitudes of these improvements consistly supass those achieved by TRC-only reinforcement. For the composite-reinforced beams, the application of prestress notably enhances the cracking load compared with the non-prestressed counterparts, while having negligible influence on the yielding and ultimate loads. The flexural stiffness of composite-reinforced beams is significantly higher than that of both unreinforced and TRC-reinforced beams. The cracks in composite-reinforced beams are characterized by their fine and dense distribution. The number and distribution of cracks in composite-reinforced beams are less affected by whether prestress is applied or not. The synergistic interaction between fiber textile and CFRP bars cn effectively reduce the stress level of longitudinal reinforcement at beam bottom. The cross-sections remain planar during bending, thus verifying the validity of plane section assumption. The calculated ultimate bearing capacity deviated by less than 10% from the measured values, demonstrating the good applicability of the proposed model. [Conclusion] The composite reinforcement can significantly improve the bearing capacity of RC beams. The proposed calculation method for bearing capacity provides an effective basis for engineering reinforcement design.

[Objective] The study investigated the main factors influencing truck fuel efficiency and their interactions, so as to improve the accuracy of fuel efficiency prediction. An interpretable truck fuel efficiency prediction model based on WOA-XGBoost and SHAP was constructed. [Method] First, the samples were reclassified according to the definition of fuel efficiency. The relevant features were computed as well based on this definition. Then, the whale optimization algorithm (WOA) and grey wolf optimizer (GWO) were used to optimize the hyperparameters of XGBoost. The models with different populations were comprehensively ranked based on evaluation metrics to identify the optimal parameters of XGBoost by using both algorithms. Subsequently, six models, i.e., WOA-XGBoost, GWO-XGBoost, XGBoost, LightGBM, Random Forest, and SVR, were used for fuel efficiency prediction. A comparative analysis on the models rankings was performed. Finally, SHAP attribution method was used to interpret the model with the highest ranking. [Result] WOA-XGBoost model has the highest comprehensive ranking. Its evaluation metrics, i.e., MSE, MAE, RMSE, R², are 0.251 2, 0.145 7, 0.501 2, 0.968 0 respectively. Cruise time and average speed are the main factors influencing fuel efficiency, with average SHAP values of 1.62 and 0.86 respectively. Cruising time and average speed are the most significant features for interaction. The average speed has a positive effect on fuel efficiency when it is over 4 km/h, otherwise it has a negative effect. [Conclusion] The proposed method demonstrates superiority in fuel efficiency prediction. The study result will provide valuable guidance for optimizing driving behavior.

[Objective] To promote the application of assembled retaining wall technology in subgrade retaining engineering, a novel assembled retaining wall structure was designed and developed. [Method] The novel structure was assembled with hollow boxes, columns and bottom plates. The boxes were filled with high-density materials for weight enhancement, and were connected with mortise-tenon joints. The columns and bottom plates were connected with tenon-pins. The boxes and columns were connected with bolts. Based on Coulomb earth pressure theory, the calculation and analysis were carried out by using Lizheng geotechnical software and ANSYS Workbench finite element simulation, i.e., the overall strength and stability of novel retaining wall, the strength and reliability of prefabricated components and connection parts. [Result] The anti-overturning and anti-sliding coefficients of novel retaining wall are 1.897 and 1.332 respectively, which meet the specification requirements of not less than 1.5 and 1.3. In the numerical simulation conditions, the maximum tensile stresses at columns, boxes and bottom plates are 0.40, 0.38, 0.69 MPa respectively; and the maximum shear stresses are 0.21, 0.47, 0.33 MPa respectively. They are all less than the standard tensile and shear strengths of C30 concrete structure. The box width and bottom plate width are the key factors influencing anti-overturning performances of the novel structure. [Conclusion] The novel retaining wall meets the requirements for anti-sliding, anti-overturning, sectional strength and load-bearing capacity. The design of assembled components and component connections are reasonable. Properly adjusting the box width or bottom plate width can effectively improve the structure stability. Compared with the existing assembled retaining walls, the novel retaining wall has obvious advantages in assembly degree, convenience of prefabrication and hoisting, reliability of connection mode, and economic benefits.

[Objective] To address the driver mind wandering during the operation of L3-level autonomous driving systems, the theoretical method for constructing a mind wandering prediction model was proposed, based on physiological indicators derived from drivers’ eye movement and heart rate data. [Method] Eye movement and ECG data were collected from drivers through simulation tests. Based on scores from the mindful attention awareness scale (MAAS), the drivers were categorized into two groups, i.e., those with a high tendency for mind wandering and those with a low tendency. The significant influencing factors contributing to mind wandering were identified to facilitate the development of a predictive model. [Result] Drivers in the high tendency group exhibited greater likelihood of experiencing mind wandering, with the frequency of such occurrences significantly increasing with longer driving durations. During episodes of mind wandering, drivers demonstrated longer gaze durations, smaller pupil diameters, slower sweep speeds, reduced sweep amplitudes, and a higher percentage of gaze directed toward driving-irrelevant areas. No significant differences were observed in heart rate variability. Utilizing driving duration, mind wandering tendencies, and five key eye movement metrics as input features, it was determined that the random forest model optimized by particle swarm optimization exhibited the best predictive performance regarding drivers’ mind wandering states. [Conclusion] The proposed model effectively identifies significant characteristics of drivers prone to mind wandering during operation. The findings provide a scientific basis for dynamic monitoring and intervention strategies aimed at mitigating driver wandering, ultimately contributing to enhanced safety in autonomous driving systems.

[Objective] The study investigated the influence of corroded steel wire accumulated multiple corrosion pits on steel wire residual fatigue life. [Method] Assuming that the steel wire corrosion pits were equivalent to the initial cracks. In view of the relative lack of experimental study results on shielding effect of multi-cracks in high strength steel wires, the fatigue experimental study on shielding effect of high strength steel wire double cracks was carried out. Double cracks were prefabricated on high strength steel wire specimens with different mutual arrangement positions, and the size variation of double cracks were considered as well. A finite element model of steel wire with double long cracks was established by using ABAQUS software, in conjunction with the commonly used fatigue life numerical simulation software FRANC3D. According to relevant crack extension parameters of cracked steel wire, the theoretical predicted fatigue life of steel wire specimen was calculated. [Result] For the specimens with same prefabricated primary crack conditions, the fatigue life of coaxial coplanar double crack specimens is lower than that of single crack specimens; the fatigue life of coaxial parallel double crack specimens is higher than that of single crack specimens; the fatigue life of specimens, with larger axial spacing between double cracks and with smaller secondary crack size, is basically the same as that of single crack specimens. The simulation analysis result indicates that the commonly used fatigue life numerical simulation software has good fit to the test results of single cracked specimens; however, the fit to the test results of double cracked specimens is relatively unsatisfactory. [Conclusion] The fatigue life of coaxial parallel double crack steel wire specimens can be evaluated by using simulation method with single crack assumption. The fatigue life of coaxial coplanar double crack steel wire specimens must be reasonably reduced based on the simulation results of single crack assumption.

[Objective] The overall bearing capacity of aluminum alloy pedestrian bridge is low,due to the low compressive stability bearing capacity of single member. To obtain axial stability bearing capacity of aluminum alloy struts, the overall instability of aluminum alloy axially compressive struts in truss structure was taken as the study object. [Method] First, the stress and deflection of strut were taken as the expedition targets. Under the condition of axial pressure iteration accuracy (0.1 kN) and strut stress iteration accuracy (0.1%), ABAQUS software and iteration program were adopted. The stability of 6061-T6 aluminum alloy axially compressive strut, with 9 kinds of I-shaped sections and 11 kinds of strut lengths, was analyzed. Then, the parametric study on 4 kinds of initial bending and 5 kinds of initial eccentricity was carried out. The relation curve of critical stress and slender ratio was obtained. The bearing capacity calculation formula with different initial defects was obtained. Finally, the results with different initial defects were compared with the current specifications. [Result] When the load is small, the initial defect has no obvious effect on the overall stability of axially compressive strut; and the effect is significant with the increase of load. According to slender ratios, the ratio curve of iterative critical stress to Euler critical stress with initial bending can be divided into 3 connected curves. According to slender ratios, the ratio curve with initial eccentricity can be divided into 2 connected curves. The ratio curves of initial eccentricity and initial bending are horizontal straight lines when the slender ratio is large. [Conclusion] In the case of small slender ratio, the calculated value with specifications takes into account the local buckling, which is safe. In the case of large slender ratio, the struts fail due to large lateral deflection, and the specifications are not considered. Therefore, the proposed calculation method is safety-oriented. The initial eccentricity, which should be strictly limited, has great influences on bearing capacity. The ultimate bearing capacity of slender strut is mostly controlled by deflection, which should be monitored in practical engineering.

[Objective] To address the issues of insufficient robustness and accuracy in vehicle state estimation by using the traditional Kalman filtering in non-Gaussian noise conditions, the unscented Kalman filtering based on the maximum correntropy criterion was proposed. This method aims to effectively suppress the influence of non-Gaussian noise, thereby significantly enhancing the reliability and accuracy of estimating key vehicle state parameters, e.g., yaw rate, longitudinal velocity, and lateral velocity. [Method] First, the nonlinear three-degree-of-freedom vehicle dynamics model was constructed. Then, based on the improved Dugoff tire model and integrating data collected by on-board sensors, a state observer capable of simultaneously observing yaw rate, longitudinal velocity, and lateral velocity was designed. Finally, through Simulink-CarSim co-simulation platform, the effectiveness of the proposed method was verified in double line and sine wave steering input conditions in non-Gaussian environment. [Result] In non-Gaussian noise conditions, the traditional unscented Kalman filtering suffers from poor convergence, weak tracking performance, and large following errors. In contrast, the proposed method effectively suppresses the non-Gaussian noise, significantly improving both convergence and tracking performance. This method can accurately and efficiently estimate the key vehicle state parameters, e.g., yaw rate, longitudinal velocity, and lateral velocity. [Conclusion] In non-Gaussian noise conditions, the proposed method demonstrates superior robustness, thereby providing more accurate and reliable state information for practical vehicle dynamics control systems, so as to effectively enhances the vehicle’s active safety and driving stability.

[Objective] To investigate the influence of measured random traffic loads on stability of single-column pier girder bridges against overturning, the statistical analysis on overturning effects of measured vehicle loads were conducted. Based on this, the study on stability against overturning based on target reliability indicators was carried out. [Method] First, the traffic flow statistics were collected by using weigh in motion (WIM) system at a bridge in Guangdong, China. Monte Carlo simulation was employed to simulate random traffic flows. Subsequently, the overturning effects of single-column pier girder bridges under random traffic loads were obtained by using the modified overturning axis method and the influence line loading method. These effects were compared with those under standard lane loads. Finally, a probabilistic model of bridge structure dead load was combined to establish the randomness model for overturning stability analysis. A method for analysis on stability against overturning based on target reliability indicators was proposed. [Result] The overturning stability coefficient of bridges calculated by using the modified overturning axis method has a relatively small error. This method can accurately assess the overturning bearing capacity of single-column pier girder bridges. The overturning effect ratio of single-column pier girder bridges is positively correlated with the number of spans and span-length. With the target reliability indicator of 5.2, the overturning stability coefficients calculated with random analysis model are 2.29 and 1.26 for curved bridges and straight bridges respectively. [Conclusion] The overturning stability coefficient of single-column pier girder bridges calculated with stochastic analysis model is smaller than that calculated with deterministic analysis model. The deterministic analysis model overestimates the overturning stability of single-column pier girder bridges, because it does not consider the randomness of load effects. The stochastic analysis model treats overturning resistance effects and overturning effects as random variables, enabling more accurate assessment of overturning stability and safety margin of single-column pier girder bridges.

Smart highway refers to a highway system that comprehensively applies new-generation information technologies and intelligent technologies to achieve digital, networked, and intelligent upgrades of highway infrastructure, thereby significantly enhancing transportation efficiency, safety, and sustainability. Although the strategic importance is increasingly prominent, current studies predominantly focus on specific technologies or regional development analyses, lacking a systematic review and comparison of the global evolutionary trajectory of smart highways. By synthesizing the trajectories of representative countries and regions, e.g., the United States, Japan, Europe, and China, the evolution of smart highways can be understood as proceeding through several historical stages, ranging from the early period of emergence and exploration, to the rise of ITS, and further to stages characterized by cooperative vehicle-infrastructure systems and digital-intelligent development. The conceptual characteristics and technological connotations of various stages are investigated. The evolutionary process of system architectures across countries is analyzed, through which a development trend is revealed toward a physical hierarchy structured around the cloud-edge-end paradigm and a logical hierarchy centered on sensing-communication-computing-application. On this basis, key enabling technologies are summarized in the domains of sensing, control, safety, and vehicle-infrastructure cooperation. The findings are expected to contribute to a more comprehensive understanding of the concepts, architectures, and technological evolution of smart highways, while providing the theoretical foundations and decision-making references for technical roadmapping, standards development, and large-scale deployment.

[Objective] This study investigated the influence of surface energy on asphalt-aggregate interfacial adhesion properties. Five types of rocks commonly used in road construction were selected as aggregate samples. [Method] The surface energy parameters of aggregates were measured with column wicking method, while the surface energy parameters of asphalt were determined with sessile drop method. The adhesion and stripping models were developed for the asphalt-aggregate system in various immersion conditions. The surface energy evaluation indices were optimized by integrating entropy weight and grey correlation methods, and associated with the results of immersion Marshall stability test and freeze-thaw splitting test as well. [Result] The higher the aggregates surface energy, the better the adhesion performance and water stability of asphalt mixtures. The relatively higher the adhesion work within asphalt-aggregate system, the stronger the water damage resistance of asphalt mixture. Van der Waals (VDW) adhesion work and spalling work demonstrate the weakest correlation with residual Marshall stability and freeze-thaw splitting strength ratio. It indicates that polar components contribute more significantly to water damage resistance than dispersive components. Furthermore, the comprehensive surface energy indicator shows high consistency with conventional water stability test result. [Conclusion] The surface energy theory can effectively and quantitatively evaluate the adhesion performance of asphalt-aggregate system, and provide theoretical basis for the evaluation of pavement resistance to water damage.

[Objective] To effectively utilize crack features, a crack segmentation method based on crack skeleton growth was proposed. The study aimed to fully retain the crack details and reduce the noise interference in pavement background, so as to improve the accuracy of pavement crack segmentation. [Method] First, the proposed algorithm preprocessed crack images by using brightness equalization method to reduce influences of shadows and uneven brightness. The segmentation threshold was obtained through empirical method to achieve preliminary segmentation of cracks. Second, based on the crack grayscale characteristics and the local directionality of crack skeleton, the preliminarily segmented crack skeletons were connected by using a set of connected domain algorithms. This process restored the inherent features of cracks, improving the recognition accuracy, while achieving good noise resistance as well. Then, after connection completion, the initial denoising was performed according to the distinguished characteristics between cracks and noise, i.e., the aspect ratio of external rectangle of connected domains, and the pixel numbers in these domains. A region-growing method, utilizing grayscale differences between seed points and their neighborhoods for growth, was proposed. Finally, by using homogeneity characteristics of crack regions, the 3×3 scale region-growing process was applied to the denoised images, making segmentation result more accurately represent the crack detail features. [Result] CrackForest dataset was selected as the test data set, and five existing crack methods were compared and verified on this data set. The result indicates that the recall rate is 87.42%, and the precision rate is 88.94%, which are better than the existing five crack segmentation methods. The F1 index value reaches 88.17%, and the precision rate and recall rate are relatively balanced, which not only achieve the high precision rate, but also ensure the stability of recall rate. [Conclusion] The proposed method achieves good detection results, not only can it complete the segmentation of obvious cracks, but also detect the surrounding small cracks. It can restore the details of cracks well, providing new ideas for the pixel-level segmentation of pavement cracks.

[Objective] The traditional earth pressure theory did not fully consider the influence of intermediate principal stress. The study systematically analyzed the influences of double-shear intermediate principal stress on earth pressure, so as to establish the calculation model. The mechanism of intermediate principal stress on earth pressure was investigated to provide accurate theoretical basis for geotechnical engineering design. [Method] Based on the unified strength theory of double-shear, six types of shear strength parameters were derived by analyzing the specific positional relation between Drucker-Prager yield criterion and Mohr-Coulomb yield criterion, which considered the intermediate principal stress influence coefficients. Furthermore, unified with Rankine’s earth pressure theory, the formulas for active and passive earth pressures incorporating double-shear intermediate principal stress were derived. [Result] Considering the influence of double-shear intermediate principal stress on both shear strength parameters and earth pressure, the active earth pressure decreases nonlinearly with the increase of intermediate principal stress influence coefficient. While the passive earth pressure increases nonlinearly as the coefficient increases. The active earth pressure result with Rankine’s earth pressure theory is larger than that with the proposed earth pressure calculation model. The passive earth pressure result with Rankine’s earth pressure theory is less than that with the proposed earth pressure calculation model. [Conclusion] The model considering intermediate principal stress can make full use of the strength potential of soil. The support structure, designed with Rankine’s passive earth pressure theory neglecting influences of intermediate principal stress, has large error and tends to be unsafe.

[Objective] The traffic flow theory can be classified into three-phase traffic flow theory and fundamental diagram approach. Due to the lack of high-precision in-situ traffic measurement, there are some controversies between two theories. A novel two-lane cellular automata model was proposed to discuss these controversies. [Method] The novel two-lane model considered velocity effect and spacing distance. When the numerical sum of proceeding vehicle virtual speed and spacing distance changes, the random slowing-down probability of the vehicle is not a fixed value, but varies with the sum. The rear vehicle tends to change the current speed and driving state when the spacing distance is large. The rear vehicle tends to maintain the current speed and driving state when the spacing distance is small. [Result] The spatio-temporal trajectory plot, at intermediate state (transition from free flow to congested flow) simulated with the proposed model, is the trajectory plot at synchronized flow state defined with three-phase traffic flow theory. The distributions of velocity and spacing distance at intermediate state indicate that there is no large-scale congested group in the road system, which satisfies the definition of interruption effect on synchronized flow state. The autocorrelation function and cross-correlation function tend to be zero, which satisfies the definition of synchronized flow state in correlation function analysis. The proposed model, consisting of simple rules and parameters, simulates the synchronized flow state. With the increase of density, the variation of lane-changing frequency has no obvious regularity before the peak value. The vehicle lane-changing frequency in the two-lane system decreases sharply after the peak value when the density increases to the critical value. When a large number of vehicles tend to slow down sharply in the road system, the free flow can directly be transited into congested flow without the transition of synchronized flow. [Conclusion] The proposed novel model focuses on the synchronized flow state and traffic phase transition between two theories that have not yet reached a consensus, and analyzes the reasonable viewpoints in both theories.

[Objective] The end of RC beam is prone to fatigue damage after being subjected to repeated actions of vehicles. It is necessary to study the fatigue shear behavior of beams with initial damage after strengthening, as well as clarify the mechanical behavior of damaged beam end after shear strengthening, thereby providing the technical support for reliable strengthening in the sheared area of beams. [Method] The side near surface mounted (SNSM) CFRP technique was adopted to strengthen RC beams. The shear fatigue mechanical behavior of beams by using this technique were studied. The material strain and the loading cross-section deflection of one comparative beam and two strengthened beams under fatigue loading were analyzed through experiments. Considering the initial damage of actual service beams before strengthening, in order to better simulate the actual situation, two strengthened beams (one strengthened only in the shear span area, and the other strengthened with longitudinal prestressed CFRP) were subjected to 2 million times of fatigue loads before undergoing shear strengthening; and then continued to withstand 500 000 times of fatigue loads. [Result] The residual shear bearing capacity of two strengthened beams increase by 5.6% and 9.8% respectively compared with the unstrengthened beams. The rapid decrease of test beam stiffness damage mainly occurs during the first 200 000 cycles of fatigue loads. The subsequent decrease of stiffness damage rate tends to be flat. In addition, due to the longitudinal prestressing of CFRP, the shear resistance of longitudinal bars is improved, resulting in better fatigue behavior of prestressed CFRP strengthened beams than ordinary CFRP strengthened beams. [Conclusion] The SNSM and longitudinal prestressed CFRP composite strengthening technique can improve the shear behavior of RC beams, and provide the technical support for shear fatigue reinforcement of beams with initial damage during service.

[Objective] To address existing traffic flow prediction methods’ insufficient mining of lane-level spatial correlation and time-dependent features, the combined Transformer-Encoder-stacked-GRU deep learning model based on lane spatio-temporal data fusion was proposed. [Method] First, the lane-level spatio-temporal features of same road sections were analyzed. The position encoding module of Transformer-Encoder framework was utilized to extract the temporal relation among input data. Second, the self-attention mechanism of Transformer-Encoder framework was adopted to complete the feature-level fusion of traffic parameters of lateral strongly correlated lanes. The spatio-temporal features about spatial correlation and temporal dependence of lateral lanes were established. Finally, the spatio-temporal features and target lane traffic parameters were taken as input of stacked-GRU module for supplement the extraction of traffic flow short-term dependent features. The model was verified by using lane-level traffic flow data of Hefei City. [Result] In the models with prediction time windows of 6 and 12, compared with stacked-GRU model, the combined Transformer-Encoder-stacked-GRU deep learning model reduces the mean squared error by 1.48% and 2.0%. Compared with stacked-LSTM model, the mean absolute error is reduced by 3.53% and 2.53%. Simultaneously, compared with the combined deep learning model considering only lateral lane spatial features, the Transformer-Encoder-stacked-GRU deep learning model with lateral spatial and multi-timescale-dependent features has better prediction accuracy. The feasibility and superiority of the proposed model in real-scene prediction are proved. [Conclusion] The proposed method can fine-graind mine lane-level spatio-temporal relation, and enhance prediction accuracy. It provides effective references for traffic management departments to evaluate traffic operation conditions.

[Objective] Construct a multi-source information fusion neural network model to improve the accuracy and real-time performance of driver behavior recognition in complex driving scenarios, as well as break through the computational efficiency bottleneck of traditional deep networks, and ensure traffic operation safety. [Method] Driver-seq, a driver behavior dataset, was constructed, including five normal driving behaviors and five bad driving behaviors. Recurrent all-pairs field transforms (RAFT) algorithm was used to preprocess image data.The interference of complex background noise was weakened by motion feature enhancement. A lightweight driver behavior recognition model based on RAFT was constructed.The optical flow estimation technique was introduced to enhance the model’s spatio-temporal feature extraction capability for dynamic driving behavior.Integrating the dual advantages of residual structure and global average pooling layer, the computational complexity was effectively controlled while guaranteeing the feature extraction depth.The GX module containing residual structure and global average pooling layer was proposed.The MobileNet-GX lightweight neural network was designed. [Result] The accuracy of driver behavior recognition reaches 98.75%. The various mainstream recognition architectures are comparatively analyzed, e.g., VGG-16, GoogLeNet, ShuffleNet-V2, DenseNet-121, MobileNet-V2.The constructed model accuracy improves by 1.74%-3.99%. The test speed improves by 6.17%-9.78%, which validates the model effectiveness. [Conclusion] Using RAFT image fusion preprocessing and MobileNet-GX deep learning model significantly improves the accuracy and real-time performance of driver behavior recognition, providing an efficient and reliable solution for intelligent driving safety monitor.

[Objective] Aiming at the high frequency of highway traffic accidents and increasingly severe traffic safety issues in urbanized areas of China, this study investigated the construction and optimization of traffic accident prediction models for multi-lane highway segments. [Method] First, the data were collected from multi-lane highways in three urbanized cites of a province, covering a total length of 1 338.74 km and 12 088 accidents over a three-year period. The dataset included information on road cross-section layouts, traffic safety facility types, traffic volumes, and accident records. Second, the collected data were processed through correlation and fusion. The traffic accident dataset for multi-lane highway segments in urbanized areas was constructed by using an equilibrium segmentation method, which excluded intersection influence based on stopping sight distance. Finally, the sectional accident prediction model was established with Random Forest, LightGBM, and XGBoost; and the model was evaluated. [Result] XGBoost outperforms other models. The parameter optimization via particle swarm optimization algorithm significantly improves XGBoost goodness-of-fit. The optimized model was analyzed by using SHAP visualization method. The result indicates that some factors have significant influence on traffic safety, e.g., width of non-motorized lanes or hard shoulders, types of separation facilities for motorized and non-motorized traffic, and traffic volume. [Conclusion] In cross-section-constrained road environments, it is recommended to moderately reduce the width of motorized lanes, widen non-motorized lanes, and prioritize the implementation of physical isolation facilities. A central median width of 1-5 m is found to be more effective in enhancing traffic safety.

[Objective] A peak particle vibration velocity calculation model considering medium interface effects was proposed to improve the accuracy of vibration response prediction on adjacent buildings during tunnel blasting. [Method] The study based on the physical mechanism of blasting stress wave propagation and reflection in rock and soil medium. The elastic wave theory was combined as well. The influences of incident angle and reflection angle on stress wave propagation path and amplitude were considered in different medium interface conversion conditions. The theoretical formula of blasting vibration peak particle velocity was proposed. The exit section of Guanlinzi tunnel on Baoji-Hanzhong expressway was taken for an engineering example. The theoretical prediction model was verified by using the monitoring data of blasting vibration velocity of in-situ monitoring points. [Result] The theoretical prediction values of peak vibration velocity were compared with the measured values of five monitoring points at monitoring section. First of all, the trend of prediction values and measured values was basically the same. Second, the peak vibration velocity decreased with the increase of distance from the blasting source. From the relative error result, the minimum error was 3.83%, the maximum error was 13.97%, and the average error was 8.21%. Therefore, the relative error was comparatively small. [Conclusion] The theoretical prediction formula of peak vibration velocity proposed in this study can accurately reflect the influence of blasting vibration on adjacent buildings. Besides, the relative error between predicted values and measured values is relatively small. The result indicates that the theoretical prediction formula considering interface effects can reasonably reflect the blasting stress wave propagation rule in different media, as well as accurately predict the peak vibration velocity.

[Objective] To effectively predict entrance slope stability risk levels for super-long mountain tunnels, the study established the entrance slope stability risk assessment model, so as to improve the scientificity and accuracy of risk assessment.[Method] CRITIC combined weighting-normal cloud evaluation model based on game theory was proposed. The key influencing factors were identified through sensitivity analysis, leading to the establishment of an evaluation indicator system encompassing topography, geological conditions, and other dimensions. Subjective and objective weights of indicators were calculated by using the optimized analytic hierarchy process (AHP) and CRITIC method respectively. Then, the weights were combinatorially optimized based on game theory. The expert scoring was integrated with statistical data characteristics by using normal cloud model and fuzzy theory. The qualitative indicators were transformed into quantitative expressions by using a forward cloud generator. The standard cloud maps were constructed to obtain the fuzzy membership matrix of each indicator. Finally, the multi-level fuzzy comprehensive evaluation was applied to determine the overall risk level. A pre-construction risk assessment was conducted on the entrance slope of super-long mountain tunnel for a highway project in Yunnan Province. [Result] The excavation method, slope structure, monitoring measurements and rock weathering degree are the dominant factors affecting stability risk. The assessed risk level is classified as Level II, indicating a relatively safe condition. The result aligns with the comparative evaluations obtained by using fuzzy comprehensive analysis, matter-element extension method, and risk matrix method. It is consistent with the actual construction feedback, thereby validating the model’s effectiveness and applicability. [Conclusion] The model integrates subjective expert judgment with the objective characteristics of evaluation data, enabling the quantification of qualitative indicators and improving the accuracy of both indicator weights and membership values. It provides meaningful guidance for classifying slope stability risk levels, and informing construction decisions for tunnel entrance slopes.

[Objective] To improve asphalt pavement performances, the basalt fiber (BF) was introduced as a modifier to composite modify two types of asphalt mixtures, i.e., AC-13C and AC-20C. For comparative analysis, the polyester fiber (PF) and glass fiber (GF) were incorporated to evaluate and contrast the pavement performances as well. [Method] Based on the raw material properties study, the study on microstructure of basalt fiber-modified asphalt was carried out. The optimum asphalt-aggregate ratio for asphalt mixture with different fiber lengths and diameters was determined through mix design. A series of laboratory tests were carried out to evaluate the influences of fiber length and monofilament diameter on pavement performances of asphalt mixture. Furthermore, by assessing both pavement performance efficiency and economic efficiency, the rational selection scheme for basalt fibers was determined. [Result] Basalt fiber exhibits adsorption and stabilizing effects on asphalt, forming 3D random distribution within asphalt matrix with interwoven and overlapping structures. With the same gradation, asphalt mixtures incorporating basalt fiber demonstrate superior overall pavement performance compared with those modified with polyester fiber or glass fiber. With the same gradation and basalt fiber length, asphalt mixtures with monofilament diameter of 17 μm exhibit better comprehensive pavement performance than those with 21 μm. With the same gradation and monofilament diameter, asphalt mixtures with larger nominal particle size require longer fibers to ensure optimum pavement performance. AC-13C mixtures with BF17-6 and AC-20C mixtures with BF17-12 achieve the optimum overall pavement performance and efficiency coefficient. [Conclusion] Considering both pavement performance and cost-effectiveness, the optimum basalt fiber specifications for AC-13C and AC-20C asphalt mixtures are determined as BF17-6(17 μm monofilament diameter, 6 mm length) and BF17-12(17 μm monofilament diameter, 12 mm length) respectively.