Firstly， the definition of mobility was discussed， and then the main existing ground vehicle mobility evaluation methods， i.e.， empirical model， semi-empirical model， numerical simulation and machine learning， were analysed and summarized comprehensively， also the advantages and disadvantages of each method were compared. In order to describe the vehicle mobility completely， application of these methods are discussed，such as military vehicles， sea-floor operation， planetary exploration and agricultural vehicles. Finally， according to the problems existing in the vehicle mobility evaluation methods， this paper proposed some key technologies and exploratory research directions from real-time evaluation of vehicle mobility and real-time terrain perception， path planning on deformed terrain and autonomous mobility evaluation for unmanned systems， so as to provide a beneficial reference to the development of vehicle mobility elevation methods.

In view of the mechanic characteristics of tunnel-type anchorage （TTA）， the research results on the clamping effect， the failure mechanism and stability， the bearing capacity and the dynamic response under seismic action are summarized. TTA can resist the huge pulling force from the main cable under the clamping effect. Based on the clamping effect， the formula of its bearing capacity can be deduced， and the mechanical parameters of surrounding rock， the wedge angle and buried depth of the plug body and other factors will have different degrees of influence on its bearing capacity. The failure of TTA starts from the cementation surface between the bottom of the plug body and surrounding rock， and develops upward in horn shape. Under earthquake， the deformation of the front anchor surface is larger than that of the rear anchor surface. Combined with the existing research contents， the future research directions of TTA mechanical properties are prospected.

First， the Hybird III dummy and CHUBM human were used to build 50 km/h frontal 100% rigid barrier virtual collision simulation platform to conduct preliminary exploratory study. Then，three finite element models of the vehicle collision were established under the Hybird III dummy seat position A and CHUBM human seat position A and B. And three cases under the driver's head， chest and thigh injuries were compared by analysis in simulation. The results showed that Hybird III dummy of seat position A and CHUBM human of seat position B had a good consistency in the dynamic response. The head and chest injuries of CHUBM human under different seat positions were higher than Hybird III dummy. The injury indicators of CHUBM human at seat position A were higher than that at seat position B， and the fracture of human ribs was occurred. The research results show that the CHUBM human has more realistic human anatomy and bionic fidelity， and has higher requirements for the safety evaluation of vehicle restraint systems， which can provide corresponding reference for the safety protection design of vehicle.

In order to accurately estimate the State of Health （SOH） of the lithium-ion battery， a method based on Douglas-Puck algorithm and Extreme Gradient Boosting （XGBoost） algorithm is proposed. Firstly， each set of voltage data is preprocessed， and the Douglas-Puck algorithm is used to vectorize the constant current charging voltage curve of each cycle. On the basis of this data， the XGBoost algorithm is applied to establish a lithium-ion battery degradation model and estimate the SOH. The results of comparative experiments show that the proposed method can effectively compress the battery voltage curve and reduce the dimension of network training data. At the same time， the developed method also has a higher prediction accuracy and faster running speed， and can realize the fast and accurate estimation of the lithium-ion battery SOH.

Considering the influence of faults between units of complex system， the relationship between the faults among the units is analyzed. Relevant parameters of associated units are obtained through Copula Numerical Interpretative Structural Modeling （CNISM）， and the fault association matrix among units is established to determine the initial clustering module. Using the degree of aggregation and coupling of modules as numerical quantization indexes， the objective function of the complex system unit fault clustering mathematical model is constructed. The Grouping Genetic Algorithm （GGA） is used to optimize the target function and determine the optimal clustering module. Finally， a certain machining center system is classified according to the above logic.

To address the dynamic behavior of luffing jib tower crane under unloading impact conditions， a virtual prototype modeling method of a crane based on a non-parametric model and a crane anti-backward-tilting test method are proposed. Firstly， through the characteristic test of the hydraulic anti-backward-tilting device （HATD） and the unloading test of the unloader， a non-parametric model describing the damping characteristics of the HATD and the excitation function of the unloading load are established respectively. Then， the rigid-flexible coupling model of the tower crane and the non-parametric model established through the unit test are integrated into the virtual prototype model of the crane， and the virtual prototype simulations under a series of unloading conditions are carried out. The remote operation method is used to implement the anti-backward-tilting test of a luffing jib tower crane， and the validity of the simulation model is verified through a series of load tests. Finally， the dynamic response of the crane structure under extreme unloading conditions is predicted through simulation， which provides a quantitative basis for the design of the key structure of the crane.

A new round pit repair method was designed to repair 5Cr5MoV cold working die steel. Microstructure in the repaired zone was characterized by scanning electron microscope（SEM）， and the microhardness distribution map of the repaired zone was drawn. a new evaluation parameter， fluctuation coefficient （n）， for evaluating the repair effect was proposed， based on the maximum hardness， minimum hardness and matrix hardness of the repaired zone. The results show that when the pit diameter is 20 mm， the repair effect is the best.

In order to explore the safety of the mainline in expressway confluence area， firstly， based on the traffic data and related research， the relationship between average speed and accident rate of the inner lane and the middle lane is analyzed by using statistical methods， then， combined with models of average speed-accident rate， Greenberg and Underwood， traffic density-accident rate models（AR-k） of inner lane and middle lane with larger and smaller traffic density k are constructed respectively， and then， influence of k of two lane and cumulative number of vehicles n on traffic safety is analyzed from the qualitative and quantitative point of view， results show that that n and k have a great influence on AR no matter in which lane and whether k is small or large. Finally， some measures to improve traffic safety are put forward from two aspects of controlling traffic volume and limiting the speed in confluence area.

To improve the problem of traffic congestion in merging area， the paper describes the lane changing characteristics and trajectories of vehicles in on-ramp area， the calculation model of minimum safe spacing for vehicles meeting the demand of vehicle merging is proposed. First, the calculation model is incorporated into the control framework of variable speed limit. The coordinated control method and model of variable speed limit in on-ramp area under the intelligent network environment are constructed by applying model predictive control method. Then，the simulation environment is built by using cellular automata to realize the speed control of connected vehicle in upstream control area. Finally,the ramp and main road vehicle arrival rate is set in 27 scenarios according to low， medium and high， and traffic volume， speed， density and travel time are selected as evaluation indexes. The traffic conditions of on-ramp area under control are analyzed through simulation. The simulation results show that the on-ramp vehicle arrival rate affects the implementation effect of the control strategy. When the density is low， the effect is not significant； when the density is medium， the advantage of the control strategy is reflected； when the density is high， the strategy is invalid.

To generate more green termination decisions that are appropriate， a fully-actuated signal timing technique for isolated signalized intersections in an environment with 100% penetration rate of connected vehicles is proposed. The concepts of green termination decision zone and vehicle extended length are defined. The lengths and number of approach lanes as well as the number， position， physical length， and speed of connected vehicles in green termination decision zone are comprehensively considered to calculate the conversion rate of vehicle extended length. On this basis， the demand of current vehicle phases is evaluated and a fully-actuated logic is designed. Regarding a typical four-leg intersection of two arterials， the simulation experiment environment is created by Vissim and Python. D-optimal design is performed to develop an experimental scheme. A total of 162，000 experimental scenarios are developed for the intersections with the new and conventional signal timing techniques， respectively. The experimental results indicate that the new technique with the recommended conversion ratio threshold of vehicle extended length has an advantage over the conventional one in reducing the average vehicle delay. The heavier the intersection-wide traffic load， the greater the advantage of the new technique over the conventional one.

When the traditional method is used to predict the traffic safety of urban underpass tunnels， the influencing factors of vehicle safety are not fully analyzed， and the prediction accuracy of traffic safety is low and the prediction results are unstable.In this paper，a safety prediction method based on edge intelligence is proposed.Firstly， an edge intelligent deep learning model is constructed to acquire and process driving data and analyze the influencing factors of traffic safety. Then， according to the analysis results， the traffic conflicts in urban underpass tunnels under different conditions of traffic flow are predicted.Finally， the predictive performance of this method is compared with other traditional machine learning by simulation.The results show that this method can predict the risk of traffic accidents， and the accuracy of traffic safety prediction is more than 90%， which is higher than other traditional methods.

Aiming at the technical difficulties of random allocation， unclear control strategy and low utilization rate of current parking guidance system， the user optimization model is established on the basis of designing user optimal strategy， the system optimization model is established on the basis of designing system optimal strategy. The period characteristic rules of the user optimization model and the system optimization model are found out. On this basis， a new bi-level optimization model is established. Verified by simulation experiments， the bi-level optimization model is better than the single model in terms of flow adaptability， the former model gets the minimum parking duration and the maximum parking turnover rate （6.53）. The research results of this paper can be applied to the parking control system of large parking lots of transportation hub.

In view of the problems of low accuracy and high requirements for equipment performance of existing target detection methods in traffic scenes， a target detection method based on improved YOLOv3 network is proposed.A new IOU（Intersection over Union） function is proposed to improve the CIOU（Complete Intersection over Union）function， as CIOU function can not accurately measure the degree of coincidence between the prediction box and the real box under special circumstances. By increasing the detection scale， the feature mAP is refined and the deep and shallow semantic information fusion is enhanced to enhance the small target detection ability； by adding SE（Squeeze-and-Excitation） attention mechanism， the model pays more attention to learning the feature information of important channels； according to the characteristics of the data set， M-YOLT enhancement method was used to improve the performance of small target detection. Through the structural channel pruning of the network， the equipment performance requirements are reduced， and Matrix NMS is used to replace NMS to improve the detection speed. The experimental results show that the proposed method can improve the detection effect and reduce the detection time. The test on KITTI data set reaches 88.4% mAP， or the average detection time of each image can be reduced to 8.6 ms while maintaining 86.3% mAP. This method can achieve better detection effect with lower hardware requirements.

Aiming at the problem of the protective capacity of the existing low-grade concrete guardrail does not match the traffic flow，a new composite guard plate which can be directly installed on the surface of concrete guardrail was proposed.A fine finite element model of vehicle-guardrail was established and the anti-collision performance of the new type combined guardrail was analyzed.Then，the protective performance of the GFRP-Concrete combined guardrail was compared with the existing concrete guardrail.The results indicate that the buffering and guiding performance of the combined guardrail is not affected， and the indexes are better than the existing concrete guardrail under the collisions of a light car and medium bus.When it was hit by a heavy truck，the concrete guardrail can not effectively suppress the roll and the vehicle will inevitably turn over.Combined guardrail can overcome to rollover and guide smoothly and the exit angle is only 0.32°.After the transformation，the protective energy of the combined guardrail reaches 290 kJ，and the protective capacity is 1.8 times than the existing low-grade guardrail.

The dynamic modulus of dense gradation asphalt mixture AC-13， stone mastic asphalt SMA-13 and six kinds of porous asphalt concrete （PAC-16， PAC-13a， PAC-13b， PAC-13c， PAC-10 and PAC-5） were tested at different temperatures and different loading frequencies. The dynamic modulus of asphalt mixture are respectively increased with the loading frequency increasing and decreased with temperature increasing， and the changing amplitudes are both gradually reduced. Compared with PAC-13b with 20.3% void content， the average values of 196.8% and 125.1% increase in dynamic modulus for SBS asphalt AC-13 and SMA-13，respectively. For the porous asphalt mixtures with closely void content， a slightly reduction of dynamic modulus is observed for porous asphalt mixture with the increase of nominal maximum particle size. When the mixtures have same the nominal maximum particle size， the dynamic modulus of porous asphalt mixture is increased about 15.8% with the reduction of void content 1%. According to the time-temperature superposition principle， the expression of the shift factor is established to describe the effect of temperature， and the master curves of different asphalt mixtures are modeled， calculated and plotted by Sigmoidal function. Various factors， such as loading frequency， testing temperature， asphalt binder viscosity， volumetric properties of asphalt mixture and aggregate gradation， are considered to develop the predictive dynamic modulus equation for different asphalt mixture. The results show that the predicted values are in good agreement with the measured results， indicating that the predicted dynamic modulus model can be applied in structure design and analysis of asphalt pavement.

In order to explore the conductive heating characteristics of conductive asphalt and mixture near the crack， conductive asphalt and mixture with different proportions were prepared and the conductive healing test was conducted. An infrared camera was used to measure the temperature field around the crack. Digital image processing technology was used to quantitatively analyze the healing level of conductive asphalt. The repeated healing properties of asphalt mixture were analyzed. Results show that the temperature of conductive asphalt and mixture both linearly increase with the charging time. There is an exponential function relationship between the heating rate and electrical resistivity. The temperature of the area near the crack of conductive asphalt and mixture is notably higher than that of others， and the local temperature rise obviously. The local high temperature is conducive to realize the target healing of cracks. The repeated healing index of asphalt mixture gradually decreases with the increasing of fracture cycles. The healing temperature and electrifying time have significant effects on the healing index. The ratio of resistivity before and after healing can be used to estimate the healing level of conductive asphalt mixture.

In the existing calculation methods of embankment retaining wall stability， the position of soil fracture surface behind the wall is assumed first， and the fracture angle θ is calculated according to the corresponding formula， then compare it with the assumed rupture angle to determine the calculation results.When the height H of the retaining wall changes， the calculation formula that the fracture surface intersects the outside， inside and middle of the load should be adopted respectively， and the obtained θ value is not unique or has no solution within a certain range of height， the hypothesis is not tenable， the specific position of the fracture surface cannot be judged， and the maximum active earth pressure of the retaining wall cannot be calculated. In view of the above problems， this paper proposes to change the vehicle load on the retaining wall from rectangular load to trapezoidal load when converting it into soil column， so as to make it closer to the nature of the soil behind the wall， and deduces a new formula for calculating the fracture angle. The value range of the angle {\phi }_{\alpha } between the trapezoidal load and the horizontal line is analyzed to make the revised formula more universal and applicable. By changing the subgrade parameters such as subgrade width， wall back inclination angle， slope height and slope rate， through comparison with the standard calculation method and actual case analysis， it is shown that the modified calculation formula has a unique solution to judge the position of the fracture surface， and the correctness and rationality of the modified formula are verified.

Combined with the basic principle of variational method and the characteristics of interaction after deformation of tied arch structure， a practical calculation method of eccentric distance increase coefficient is proposed. Through engineering case analysis， the calculation result of this paper is compared with the normative method of bridge and the finite element method. The results show that the calculation formula specified in the code has the largest result， the calculation result of the formula is large， the finite element calculation result is small. This above means the proposed method is widely applicable and safe. The calculations of the eccentricity increase coefficient of are quite different in catenary， parabola， circular arc and straight rod with constraints at both ends， related to the constraint conditions of the tie-rod arch and the forms of load.

In order to address the problem of slow training of the current deep ResNets model, a novel residual structure is designed. Compared with the typical residual structure， the structure only contains a Batch Normalization and ReLU， which reduces training time and improves the training speed by reducing the amount of calculation in the network training process. The comparative experiments are carried out on the CIFAR10/100 image classification database. The classification error rate of 110 layers networks constructed by this method on CIFAR10 and CIFAR100 is 5.29% and 24.80%， respectively. The classification error rate of 110-ResNet is 5.75% and 26.02%， respectively. Training the network takes 133.47 （this method） and 208.26 （ResNet） seconds per epoch， increased by 35.91%. The results show that the network structure greatly improves the training speed while ensuring the classification performance， and has better practical value.

In order to discover abnormal data in the data stream in time and reduce potential threats to the network， a high-dimensional category attribute data stream outlier mining algorithm based on spectral clustering is proposed. The characteristics of orderliness, high speed and high dimensionality of data streams are analyzed, and the main sources of outliers are explored.Using the attribute weight quantization method, introducing information entropy, merging the data streams with strong relevance, and then reducing the dimensionality of the data streams to reduce interference. The spectral clustering algorithm is used to set key scale parameters, the distance between the sample and the target is calculated by the affinity matrix, the spectral clustering is transformed into an undirected graph segmentation problem, the feature matrix is obtained, and the significant outlier features are extracted.Using the distance mining method, data blocks is added to the data stream, the probability distribution between two adjacent data blocks is judged, a sliding window is set, the distance between the data and the sliding window is obtained, and then compare with the set threshold. Outliers are added to the set to complete the mining.

The simulation results show that for data streams of different sizes and dimensions, the execution time required by the algorithm is within 42 s and 40 s respectively, and it has good scalability for the size and dimensions of data streams, and the outlier data mined is consistent with the reality.

Aiming at the Yungang Grottoes corpus with sparse data， the completion of sparse data is realized through the artificial understanding of semantic rules， which affects the accuracy of the follow-up work to a certain extent. Therefore， an automatic completion algorithm for missing connection of knowledge atlas considering data sparsity is proposed， and some corresponding contents are also proposed in the text. By setting the neighborhood structure of the data， the knowledge graph embedded representation model based on data sparsity is constructed to extract the unknown relationship of the data. Then， the long-short-term memory network model is used to automatically complete the missing data in the variable-length sequence， and the automatic completion of the missing connection of the knowledge graph is realized. Finally， the algorithm is applied to the construction of the knowledge graph of Yungang Grottoes, it can be seen from the experimental results that on the same database set， considering the data sparsity， the accuracy of the algorithm is up to 95.4%，much higher than other traditional algorithms.

To address the problem that the Ball k-means （BKM） algorithm is sensitive to the initialized centers and the solution accuracy is not high， an incremental k-means clustering algorithm based on multi-ball splitting is proposed. The algorithm takes advantage of the feature that BKM needs to record the clustering radius of each ball， starts from a given number of initial clustering centers， and generates multiple incremental clustering centers at once according to a fixed step until k initial centers are obtained. The algorithm avoids the calculation of distances between all n s-dimensional data and reduces the computation of traditional incremental clustering center selection from O(n^{\mathrm{2}}s) to O\left(k\mathrm{l}\mathrm{o}\mathrm{g}k\right) （k<n） without increasing the space complexity. The experimental results of comparison with k-means， BKM， IK-+and three typical incremental clustering algorithms on 8 UCI datasets show that the proposed algorithm reduces the initialization sensitivity of BKM and improves the solution accuracy by 37.15%~66.92%； it is the only algorithm among the comparative incremental algorithms that can improve the efficiency of k-means， and the efficiency advantage becomes more obvious as the number of clusters increases.

In order to deal with the environmental changes of dynamic multi-objective optimization more effectively， a dynamic multi-objective optimization algorithm based on Kalman filter prediction strategy is proposed. In the evolution process， a new calculation method is used to calculate the population center point. When the environment changes， the Kalman filter prediction model is used to predict the current population center point， and the approximate true Pareto center point of optimal solution set is used to correct the prediction value， and new individuals are generated based on the modified center point to reinitialize the population during the running of the algorithm； In order to increase the diversity of the population， five new individuals are randomly generated from the search space during the operation of the algorithm， and the corresponding number of individuals in the current population are randomly replaced. Compared with other dynamic multi-objective optimization algorithms in multiple test functions， the results show that the value of Modified Inverted Generational Distance （MIGD） in the whole evolution process is relatively small， and the value of Inverted Generational Distance （IGD） in the evolution is generally smaller than that of the contrast algorithm， and the calculation time is equivalent to that of the comparison algorithm.

In order to obtain better angle estimation performance at less cost， a beamspace conjugate ESPRIT algorithm for direction of arrival （DOA） estimation in multiple-input multiple-output （MIMO） radar is presented. Firstly， the reduced-dimensional transformation is utilized to transform the observed data into a lower-dimensional space. Then an augmented observation data matrix with double number of effective elements is constructed by taking advantage of the property of the noncircular signals and is mapped into beamspace. Finally， the rotational invariance property of the real-valued beamspace array manifold is constructed and exploited to find the directions of the targets. Compared to the conventional ESPRIT， the presented method can obtain greatly improved estimation performance and can achieve a significant reduction in the amount of computation. Numerical examples are presented to illustrate the performance of the presented algorithm.

To make a reasonable cooperation mode between nodes， reduce the activation time of nodes， balance the energy consumption between nodes and prolong the network lifetime while ensuring the target tracking accuracy， a nodes scheduling algorithm is formed based on the residual energy and PCRLB information of nodes in the target tracking system of wireless sensor network. First, the proposed algorithm uses particle filter to predict the position of the target at the next time. Based on the predicted position and the prediction covariance matrix， the ellipse area of the target at the next time is determined， and then the candidate nodes region at the next time is determined. Then,in the candidate nodes region， based on the information of PCRLB and the residual energy of nodes， several sensor nodes are selected to form a cluster to track the target. Simulation results show that the proposed algorithm can not only reduce the tracking error， but also balance the energy consumption of nodes and prolong the network lifetime.