Based on the fixed-volume bomb test platform and the fixed-volume bomb simulation model constructed by CONVERGE software， the influence of the nozzle structure（nozzle diameter， angle between holes， and number of holes） on the characteristic parameters of the hydrogen jet is investigated. It is found that the penetration distance， area， volume and volume-absorption ratio of the hydrogen jet first increase and then decrease when the diameter of the jet hole increases. The initial cone angle of the jet is larger for small apertures， the cone angle of the jet tends to stabilize for all apertures； when the cone angle between the jets is increased， the jet penetration distance， area， volume， and convolution-absorption ratio decrease and then increase， while the jet cone angle increases and then decreases； when the number of the jets is increased， the jet penetration distance， area， volume， and convolution-absorption ratio decrease and then increase， and the stabilized cone angle of the jet with multiple apertures is even larger. The results reveal the relationship between the structure of the nozzle holes and the characteristics of the hydrogen jet， which provides an important basis for the optimization of the hydrogen injection system.

This study aims to enhance the feedback efficiency and perceptual consistency of in-vehicle interaction systems under complex driving conditions by developing a pseudo-haptic signal system that integrates visual， auditory， and tactile modalities. The system is built using a capacitive touchscreen and piezoelectric actuators， with a designed vibration parameter matrix to model the mapping between perceptual attributes and vibration features. Kriging interpolation is employed to construct three-dimensional perceptual surfaces. Two signal generation methods—frequency mapping and harmonic envelope modulation—are proposed， with the latter showing better performance in frequency alignment and signal discriminability. A recognition experiment was conducted using four levels of stimulus quantity. Results indicate that the visual modality achieved the highest identification performance， while the tactile modality was more sensitive to changes in stimulus number. A 3.2 km driving simulation scenario was constructed to represent typical driving conditions. Under this setup， the auditory-tactile concurrent masking configuration（A+H（+V）） yielded the lowest RMS and peak yaw rates across most conditions， indicating the highest driving stability under this secondary task， along with the best overall performance in terms of response time and subjective evaluation.

To achieve accurate prediction of surrounding vehicles' future trajectories for autonomous vehicles under single-vehicle intelligence， a trajectory prediction method integrating vehicle perception bounding boxes and motion states is proposed. By fusing data from onboard camera images， LiDAR， and ego-vehicle motion information， the proposed method extensively extracts the motion features of target vehicles and the motion trends of the ego vehicle. First， the target vehicle's trajectory and displacement are encoded， and a cross-modal attention mechanism with visual feature fusion is established. Second， an ego-vehicle displacement-difference/angle-difference encoder is designed； combined with optical flow features， it accurately captures the ego vehicle's motion trend， and a Transformer architecture is employed to achieve deep fusion of the target vehicle's motion features with the ego vehicle's motion features. Finally， the fused features， together with the target vehicle's optical flow features， are fed into a decoder to generate the predicted trajectory of the target vehicle. The proposed trajectory prediction model was trained and tested on the NuScenes dataset and deployed on a modified Dongfeng AX7 intelligent vehicle for real-vehicle experiments. The results demonstrate that the proposed method has advantages in both prediction accuracy and deployment capability.

To solve the problem of thermal deformation in the surface dressing of Φ1 200 mm metal polishing disc during the continuous polishing process of large aperture planar optical elements， a compensation method of thermal error in the surface dressing of metal polishing pad using fluid pressure is proposed. The structure of fluid pressure control system on the surface of Φ1 200 mm metal polishing pad was designed. Three aspects of the experiment were studied， including the thermal deformation of the polishing pad surface， the fluid pressure control deformation of the polishing pad surface， and the thermal error compensation of the polishing pad surface. The experimental results showed that the surface thermal error compensation of polishing pad can be realized by adjusting the fluid pressure. When liquid inlet pressure was 0.10 MPa， the best effect of surface thermal error compensation of Φ1 200 mm polishing pad was obtained， in which the error peak and valley value was reduced from the original 24 μm to 14 μm， and the thermal error was reduced by 42%.

Aiming at the model mismatch problem existing in the application of kinematic model to trajectory tracking control of unmanned tracked vehicles， a trajectory tracking control method based on feedforward compensation of dynamics was proposed for tracked vehicles. Firstly， based on the shear force-displacement theory， a steering dynamics model of the tracked vehicle was established， and the effectiveness of the dynamics model was verified. Then， taking a zero-differential steering unmanned tracked vehicle as the research object， a trajectory tracking error model was established， and a model predictive controller with dynamic feedforward compensation was designed. Finally， the proposed method was verified through field vehicle tests. The experimental results show that after the trajectory tracking controller based on the kinematic model is compensated by dynamic feedforward， the average lateral error and average heading angle error of the unmanned tracked vehicle's trajectory tracking control are reduced by 55.67% and 24.25%， respectively. The proposed method can be used as a reference for improving the trajectory tracking control accuracy of unmanned tracked vehicles.

Considering the impact of low visibility on the safety distance， acceleration， sight distance， and lane-changing behavior of human-driven vehicles （HDV）， a car-following model for HDV is developed based on cellular automata， while the car-following behavior of connected and automated vehicles （CAV） is characterized using a model developed by the Path Laboratory. A lane-changing model for HDV is constructed， incorporating factors such as visibility and driving style， whereas CAV lane-changing decisions are modeled based on the type of preceding vehicle and lane-changing conditions. Simulation studies on heterogeneous traffic flow under low visibility conditions are conducted， and the simulation results are used to evaluate capacity， congestion rate， conflict severity， average fuel consumption， and emissions under different visibility levels and CAV penetration rates.The simulation results demonstrate that as visibility and CAV penetration rates increase， both capacity and average speed significantly improve. Compared to Nasch models， the proposed car-following and lane-changing models for heterogeneous traffic flow under low visibility conditions enhance safety by approximately 23.1%， while reducing average fuel consumption and nitrogen oxide emissions by 3.42% and 1.5%， respectively. This study enriches the theoretical research on highway traffic flow and provides a foundational model for investigating traffic-related issues in heterogeneous traffic environments.

To clarify the key influencing factors of vehicle interaction risk characteristics and severity in the short weaving area of urban expressways， the Dashuying Interchange in Kunming City was taken as the research object. Based on the trajectory data extracted from drone aerial videos， a dataset of two vehicle interaction risk values was calculated through a driving risk field model. The cumulative frequency method was used to divide the interaction risk into four levels of different risk severity. The results showed that severe risks were mainly distributed in the upper half of lane 1， lane 3， and lane 4. Selecting 11 key factors that affect the severity of interaction risk as feature variables， XGBoost and RF algorithms were used to construct interaction risk severity prediction models. The results showed that the accuracy of XGBoost algorithm prediction was 92.92%， and the F₁ score was 92.91%. The accuracy of RF algorithm prediction is 90.22%， and the F₁ score is 90.19%. To further reveal the mechanism of action of feature variables， SHAP was introduced to quantify the contribution of 11 feature variables to the severity of interaction risk. The results showed that longitudinal spacing， lateral spacing， target vehicle speed， and two vehicle speed difference were the top four variables in terms of contribution. The research results can provide a basis for active vehicle prevention and control and traffic safety management in short weaving areas of expressways.

To address the issues of resource waste and operational inefficiency faced by fixed-capacity buses serving as feeders for urban rail transit systems， a dynamic scheduling method for modular bus main routes was proposed. By designing a decoupling-coupling point determination mechanism and a dynamic dispatching mode， a scheduling optimization model was established with the objectives of minimizing vehicle procurement costs， operational costs， and passenger time costs. The model was solved using an Improved Sparrow Search Algorithm （ISSA）. A case study was conducted in the service area of the Dongwu Boulevard Station on Wuhan Metro Line 1. The DBSCAN algorithm was applied to identify shared stops， and the A^* algorithm was utilized to plan the main feeder route. Results demonstrate that compared to the fixed bus operation mode and the independent operation mode， the modular bus main route operation mode reduces the total system cost by 11.59% and 4.54% respectively， and increases the seat utilization rate by 23.45% and 11.50% respectively， which can improve the resource allocation of public transport to a certain extent.

To investigate the nonlinear relationship between the built environment and pedestrian volume， based on street view imagery from Nanjing's main urban area， pedestrian volume and streetscape features were extracted using deep learning. Multi-source data were integrated to construct built environment indicators at both the neighborhood and street levels. An Extreme Gradient Boosting （XGBoost） model identified key factors in the built environment and characterized their nonlinear associations with pedestrian volume. The results reveal that among the neighborhood-level built environment factors， distance from the city center， the number of service facilities， and the number of restaurants significantly affect pedestrian volume； whereas among the street-level built environment factors， terrain， sky index， and green view index contribute more substantially. Additionally， all variables exhibit nonlinear associations with distinct interaction patterns. The impact of distance from the city center and terrain on pedestrian volume tends to be negative； in contrast， the interaction patterns. The impact of distance from the city center and terrain on pedestrian volume tends to be negative. In contrast， the number of service facilities， the number of restaurants， and the green view index are positively associated with pedestrian volume. Moreover， the nonlinear relationship between sky index and pedestrian volume presents an inverted “U”-shaped pattern. The research findings presented can serve as a theoretical reference for optimizing urban spatial layout and improving pedestrian-friendly transportation policies.

In order to improve the lag response of traditional signal control methods to real-time fluctuations of traffic flow， a joint optimization method of intersection signal phase and vehicle trajectory in a connected environment was proposed. By collaboratively optimizing the vehicle trajectory and signal phase timing scheme， it can dynamically adapt to traffic flow fluctuations and achieve the purpose of improving the traffic efficiency of intersections. The simulation results show that compared with the signal control strategy without trajectory optimization， the proposed method can reduce the average vehicle delay by 6.45%~16.42% and the parking rate by 2.36%~8.39%. In the scenario of large random fluctuation of traffic flow， the performance of the collaborative optimization method is more stable than that of the mixed integer linear programming. Further analysis shows that the higher the penetration rate of connected vehicles， the more significant the control effectiveness of the proposed method.

To elucidate the complex mechanisms underlying residents' travel mode choice under urban traffic congestion and to optimize the modal composition of public transport and private-car travel， a multidimensional-cost-based method for travel mode choice modeling and regulation was proposed. By integrating individual daily travel records， 8 368 daily trip chains were generated， and a multinomial Logit model incorporating socioeconomic attributes was developed.Results showed that expenses costs and time costs significantly influenced mode selection. Subsequently， 13 806 filtered travel records were fused with multimodal spatiotemporal data from Baidu Map API to establish cost scenarios. Comparative analysis of machine learning models identified XGBoost as the optimal predictor. Cost regulation simulations demonstrated that reducing public transit costs by 20% led to a significant decline in the proportion of private car usage， with a decrease of 4.65 percentage points. These findings validate the efficacy of cost-driven policy tools in steering travel behavior， providing strategies for differentiated transportation pricing and advancing green urban mobility transitions.

Based on one-year， second-by-second on-board diagnostic（OBD） monitoring data from 10 light-duty vehicles in Xi'an， road slopes were extracted by combining high-resolution digital elevation models（DEMs） with geographic information system（GIS） path-matching techniques. Vehicle-specific power（VSP） theory and the Motor Vehicle Emission Simulator（MOVES） model were employed to estimate emissions of carbon monoxide（CO）， hydrocarbons（HC）， nitrogen oxides（NO _x ）， and particulate matter（PM）. To address the limitations of conventional models in capturing the coupled effects of speed and slope， a bivariate polynomial regression model of speed-slope was developed， and an uphill-downhill partitioning strategy was applied to analyze the nonlinear characteristics of emissions. Results show that the proposed model achieves coefficient of determination（R²） values exceeding 0.82， with favorable error metrics. Specifically， emissions of all pollutants increase significantly with speed and slope under uphill conditions， whereas downhill operations exhibit generally lower emissions but with local peak values. Moreover， three-dimensional response surfaces indicate that HC and NO _x are more sensitive to variations in operating conditions. This study provides a high-resolution quantification of the speed-slope-emission relationship， offering theoretical and empirical support for accurate urban micro-scale emission estimation， driving condition identification， and traffic-related emission reduction strategies.

To investigate the seismic response characteristics of ballastless tracks on multi-span simply supported bridges in high-speed railways located in large-temperature-difference regions of Northwest China， this study takes a 10-span simply supported bridge with CRTS Ⅲ slab ballastless track as the research object. An integrated track-bridge finite element model is established， and dynamic time-history analysis is employed to systematically examine the seismic response patterns of track components and interlayer connections under coupled environmental temperature and seismic effects. The results indicate that： ① Under seismic conditions with PGA≤0.2g（g=9.81 m/s2） and temperatures ranging from -30℃ to 40℃， the longitudinal stress of the rail remains within a safe range， with low-temperature conditions exerting a more pronounced influence than high temperatures under the same seismic intensity. ② Compared to minor earthquakes， structural displacement responses are more sensitive to temperature variations—high temperatures significantly increase longitudinal displacements， while low temperatures lead to a more notable rise in rail vertical displacements. ③ Under coupled temperature-seismic effects， the peak longitudinal deformation between the rail and track slab occurs at the end spans， yet exhibits significant spatial variation with temperature： at -50 ℃， deformation in the first span increases by 131.83% compared to normal temperature， whereas at 50 ℃， the tenth span experiences a 137.54% increase. ④ Under minor-to-moderate earthquakes， the track structure demonstrates optimal seismic performance at -20 ℃， with all response parameters maintained at relatively low levels.

Multi-sensor fusion is crucial for achieving accurate and reliable autonomous driving. However， current point-level fusion methods enhance LiDAR point cloud features by projecting camera features onto LiDAR data， a process that often discards semantic information from camera features due to projection errors. To address the loss of local semantic information in images and the introduction of excessive background points during keypoint sampling， we propose an enhanced algorithm based on the BEVFusion network （a simple and robust LiDAR-Camera fusion framework）. Specifically， dilated convolutions and a weighted bidirectional feature pyramid are introduced in the image stream to expand the perceptual field while preserving semantic integrity. Additionally， a semantic-aware farthest point sampling （SA-FPS） mechanism is proposed to prioritize foreground points with richer information. Building on the dynamic fusion framework， we integrate squeeze-and-excitation （SE） operations and design an adaptive gated dual-dynamic fusion network to dynamically select valuable information from different modalities. Experimental evaluations on the KITTI dataset and generalization tests on the nuScenes dataset demonstrate that our algorithm achieves superior detection accuracy compared to state-of-the-art methods.

Although existing tensor-based multi-view clustering algorithms have achieved good clustering results， they have certain limitations：① These methods often ignore the unique local geometric structure of each view， making it difficult to cluster data with nonlinear structures； ② They are highly susceptible to large-scale and high-intensity noise； ③ The tensor nuclear norm， typically based on singular value decomposition， applies a uniform shrinkage strategy to singular values， failing to account for the intrinsic differences between them. To solve these problems， a robust multi-view clustering algorithm based on integrating high-order information and local structure is proposed. The algorithm integrates the weighted tensor Schatten-p norm and hyper-Laplacian regularization to capture high-order information while retaining the unique local geometric structure of each view， solving the clustering problem for nonlinear data. At the same time， l_1，2 norm is used to enhance the cluster structure， and Cauchy pseudo-norm is used to constrain the noise matrix to reduce the impact of noisy data on the clustering results. Experimental results with 10 comparison algorithms on 8 publicly available multi-view datasets show that the performance of the proposed algorithm is better than the comparison algorithm in most cases.

To tackle the issues of inadequate focusing accuracy and poor convergence performance exhibited by traditional phased-array ultrasonic focusing algorithms in complex environments with multi-physics coupling， this paper introduces an improved Equilibrium Optimizer algorithm. The proposed method incorporates a main-sidelobe-separated objective function， a hybrid particle initialization strategy based on Cauchy distribution and random sampling， and a Sigmoid-based variable generation probability mechanism. These enhancements effectively address problems such as ambiguous focal field distribution， non-uniform initial particle dispersion， and insufficient exploration-exploitation balance. Simulation results demonstrate that， compared with the original EO， the proposed algorithm achieves a 23% increase in convergence speed and a 4.8% improvement in acoustic pressure optimization. Moreover， it exhibits significant advantages over other mainstream algorithms in both convergence efficiency and acoustic pressure optimization performance.

To address the decline in detection accuracy of distant and small-scale objects in sparse point cloud environments due to limited feature extraction capabilities， this paper proposes a novel 3D object detection algorithm， CAResPillars， based on multiple attention mechanisms and deep residual networks. Built upon the PointPillars framework， the method introduces a Point-Channel Attention （PC-Attention） mechanism in the point cloud feature encoding stage. This dual attention design， operating at both point andchannel levels， adaptively enhances critical geometric features to mitigate the feature insufficiency problem in sparse point clouds.In the BEV feature extraction stage， a deep residual backbone network named CAResNet42， integrated with a coordinate attention mechanism， is employed to strengthen the model's ability to capture features of long-range and small-scale objects while addressing the vanishing gradient issue in deep networks. Experiments conducted on the KITTI dataset demonstrate that CAResPillars achieves notable performance gains over the baseline. In particular， the mean Average Precision （mAP） for pedestrian detection， typically involving smaller 3D objects， improves by up to 6.39%， with vehicle detection accuracy also showing consistent improvement. Ablation studies confirm the effectiveness of the proposed modules， and real-world deployment validates the algorithm's robustness and applicability in complex environments. The proposed method offers an effective and practical solution for real-world 3D object detection tasks， highlighting its potential industrial value.

Semi-supervised learning serves as a core technique to address the scarcity of labeled data in medical image segmentation. However， existing methods struggle to achieve satisfactory segmentation results for complex medical images with high morphological heterogeneity and blurred boundaries. This paper proposes a semi-supervised segmentation algorithm based on Dual-Scale Entangled Complementary Learning. A Dual-Scale Entangled Complementary Decoder is constructed to establish collaborative complementary relationships between dual-scale features and realize complementary modeling of structural information. Specifically， the Cross-level Fusion Encoder Module strengthens the interaction between shallow detailed features and deep semantic features， while the Multi-Dimensional Feature Optimization Module improves the reconstruction capability of spatial structures and boundary details during decoding. Supervised loss together with perturbation and scale consistency losses are adopted to implement semi-supervised training. Experimental results demonstrate that the proposed algorithm achieves competitive segmentation performance on datasets of colon polyps， skin lesions and brain tumors， and maintains favorable stability in segmentation accuracy， boundary smoothness and cross-test-set generalization ability.

This paper proposes a green wireless system， base stations are powered by renewable power sources （RPS） to reduce carbon dioxide emissions and improve efficiency in new wireless networks. The proposed method comprises the capacity-limited RPS and the Base Station（BS）. The BS can meet random connection demands of mobile terminals. The BS， powered by a combination of RPS and power grid， can monitor the RPS. Cooperation among RPSs is essential for implementing distribution strategies to supply power. The BS can reserve power from RPS， and the RPS determines the energy allocated to the BS. The paper solves the issue as a non-cooperative game between BSs and RPSs to denote the relationships among individuals. Firstly， the study adopts an M/M/1 queuing model to analyze the transition and investigate the decentralized game； it analyzes the competitive relationship between the reserve strategy and supply rate and mathematically describes a Nash Equilibrium （NE） of a non-cooperative game to find the causes of inefficiencies. A set of easy linear contracts is used to coordinate the system. Secondly， it employs an Incentive Compatible （IC） method to accurately report demands. Finally， simulational results indicate that， in scenarios with low link requirements， the solution can significantly reduce the consumption of traditional energy sources for base stations； and even if traffic is heavy， the proposed scheme can still reduce energy costs by integrating renewable energy alongside conventional energy through renewable energy BSs.