To solve the problems that the use conditions of hybrid electric vehicle power batteries deteriorate and the full-life-cycle cost of vehicle increases due to complex driving cycle， a planetary hybrid electric vehicle was taken as the research object to proactively extend the battery life from the perspective of energy management optimization. Based on the historical driving data， a representative real vehicle driving cycle was built. The dynamic programming algorithm was used to solve the multi-objective optimization problem with the minimum overall fuel consumption and battery life attenuation to ensure the global optimal system performance. Due to the problem that dynamic programming algorithm has limitation of driving cycle and demands large amount of calculation， the neural network controller was trained for realizing energy management control based on the global optimization results. Simulation results show that compared with the optimization with a single goal of fuel consumption， multi-objective optimization can reduce battery life attenuation by 43.28% and increase fuel consumption by only 1.22%， which slows battery life attenuation while taking into account fuel economy. The control strategy based on neural network achieves optimized control that actively adapts to the daily driving cycle of the driver， which has a good application prospect.

To better understand the mechanisms and risks of abdominal injuries that result from underbody blast events， a finite element model including the occupant， the vehicle structure and the flow field of explosion was established. The dynamic responses of the thoracic and abdominal organs under the explosion of 6 kg TNT charge were studied， and the effects of seat impact， floor impact and seat belt loading on abdominal pressure were compared. Injury risks of the abdomen during blast loading were evaluated based on two biomechanical predictors： strain energy density of parenchymatous organ and the peak rate of abdominal vascular pressure change. During the process of impact， the occupant was pushed upward by the seat impact， which made the abdomen expanded. When the occupant was separated from the seat， the seat belt restricted the occupant from moving upward， which compressed the abdomen. The simulation results show that the seat impact has the largest effect on the abdominal pressure， followed by the seat belt loading. In the model constructed in this paper， the peak rate of abdominal vascular pressure change was 2.1 kPa/ms which showed that the risk of abdominal injury during blast is low. The present results can be served as a reference for vehicle protection system design.

A man-machine shared driving model for the intelligent vehicle was proposed， employing human drivers' real-time response to the environmental risk. Firstly， typical driving segments， including car-following and cut-in segments， were extracted from a real-traffic-based dataset， the highD dataset. Then the driving risk field theory was employed to quantify the environmental risk in extracted driving segments. By fitting the environmental risk effect and driving acceleration， a safe risk-response strategy was obtained， following which the Flexible Control-Transition Model （FCTM） based on strategy deviation was proposed. Finally， the Longitudinal Control Model （LCM） was applied as the auxiliary control model， and the man-machine shared driving simulation was carried out in two dangerous driving scenes， namely front-vehicle emergency braking and adjacent-car cut-in. The results show that the proposed FCTM can modify driving behavior of the human driver in dangerous scenarios through smooth man-machine control transition and improve driving safety.

Firstly， linear and nonlinear multi-freedom torsional vibration models of the electrified powertrain are established respectively which contains a clutchless two-speed Automated Mechanical Transmission （AMT）. The sliding sleeve's speed fluctuation and dynamic behaviors of the two models at different phases of shift process are analyzed. The differences of shift quality between the two models are compared. Secondly， a time delay Proportional-Integral-Derivative （PID） control method of speed difference at the motor speed phase is proposed to improve shift quality. The results show that the nonlinear torsional vibration of gears extends shift time， increases shift impact and friction work of synchronizer. Appropriate time delay can decrease shift time and minish impact peak and friction work.

The drilling tool conveying mechanism is a rigid-flexible coupling multibody system. Due to the uncertainty of system parameters and external disturbances during operation， its working stability is reduced. To solve this problem， a nonlinear dynamic model of the drilling tool conveying mechanism was established using the multibody system transfer matrix method， and a trajectory function was set according to the work requirements. The working stability control of the drilling tool conveying mechanism is divided into two parts： working position and attitude tracking control and nonlinear vibration suppression of the flexible body in the mechanism. A PID control algorithm based on BP neural network is proposed to dynamically track the posture of the whole mechanism. For the vibration of the flexible body， a piezoelectric material is pasted on the surface of the flexible member to form an intelligent composite material， and the sliding mode control algorithm based on fuzzy RBF neural network is used for suppression. The simulation results show that the working position accuracy of the whole mechanism is significantly improved， and the vibration of the flexible body is effectively suppressed.

Under high-pressure and high-speed working condition， double-acting vane pumps are prone to cavitation near damping grooves of the high-pressure oil outlet. In severe cases， cavitation could destroy surface of the valve plate， bring sharp noise and decreasing the service life. In order to figure out the cavitation mechanism， the Computational Fluid Dynamics （CFD） simulation method was used to model and analyze the internal flow field in a double-acting vane pump. The dynamic grid technique was applied to define the boundary motion of the rotor and rollers through the user-defined function （UDF）. The fluid dynamic characteristics of velocity field and pressure field were calculated and analyzed under different rotational speed and system pressure. Furthermore， the occurrence of the cavitation was analyzed. Because of the existence of a local low-pressure flow field inside the oil gap between the rotor and valve plate， the cavitation occurred in front of the triangular damping groove of the high-pressure outlet cavity. Comparing the simulation results with the experiment ones， it shows good accuracy and reliability.

A method for checking the bonding strength of high-speed Electric Multiple Units （EMU） side window based on residual strength is proposed from the application point of engineering considerations. This method enables to check the bonding strength of the adhesive layer by combination of accelerated aging experiments and simulation analysis. Using the implementation of the program to complete the test results of the display， obtaining the safety margin cloud map of adhesive layer， which is convenient for the engineering staff. Finally， the effectiveness of the method is verified by the actual vehicle model.

The effects of thermal deformation on the failure strain， ultimate stress， and yield stress of AA5754 aluminum alloy are analyzed by high temperature pre-tension test and room temperature tensile test. A damage coupling pre-forming constitutive model is established to characterize the relationship between forming strain and service performance. This constitutive model is used to analyze the effects of forming damage and thickness change on the bending performance of side-door impact beam. The results show that forming damage directly reduces the service performance of AA5754 aluminum alloy. When the forming strain is 0.7 at 300 ℃， the failure strain and ultimate stress are reduced by 62% and 16%， respectively. The proposed constitutive model can accurately predict the effect of hot forming on the static mechanical properties of AA5754 aluminum alloy.

The parameter system of the Cellular Automata Finite Element method （CAFE） for simulating ductile iron grain structure was optimized. Based on a large number of preliminary experiments， the surface nucleation parameters and the nucleus parameters of the eutectic grains were optimized. The flow length of the spheroidal graphite spiral samples with various carbon and silicon contents and the solidification structure at the tip of the stream were simulated. In the verification test， the grain boundary model was realized by the constitutive algorithm for the inconspicuous ductile iron eutectic grains. The flow length of the obtained spiral sample and Voronoi grain distribution map were compared with the simulation results. It was shown that after the carbonization or silicon enhancement treatment， the crystal grains at the end of the spiral sample stream were refined in the hypoeutectic ductile iron sample， and the fluidity of the sample was improved. Moreover， the simulation results are in good agreement with the experimental verification results， and the method for evaluating the fluidity by numerical simulation is established.

In order to reveal the characteristics of rock breaking process with double bits' self-balancing and synchronous reverse rotation. In this paper， through theoretical calculation， numerical simulation and hole wall morphology analysis， the rock fragmentation characteristics of dual impregnated diamond bit and single impregnated diamond bit are compared and studied. The simulation data show that show that that the double bits present bidirectional small step double screw drilling at the bottom of the hole， while the single bit presents unidirectional small step spiral drilling； due to the reverse rotation of the inner and outer bits of the double impregnated diamond bit， the rock fragmentation characteristics of the single impregnated diamond bit are compared The experiment data show that the torque balance effect on the bottom rock is smaller， the disturbance to the formation is weaker， and the hole wall is more smooth and stable； the double bit produces more stress concentration points for the rock at the bottom of the hole， the maximum stress value is 114.7% of the single bit， and the axial displacement is 121.7% of the single bit， which indicates that the double bit is more likely to break the rock. The research results provide theoretical support for the research of double bit self balanced drilling.

The aims of this study are to achieve a quantitative assessment of the risk of roadside accidents on highways and to propose corresponding safety measures to reduce accident losses. First， the acceleration severity index （ASI） is used as the indicator of occupant injuries， and the horizontal radii， vehicle departure speeds， side slope and subgrade height are taken as research variables in this research. Second， collision tests of trucks and cars were carried out in the presence or absence of roadside guardrails by constructing the vehicle， road and guardrail models in PC-crash simulation software， a total of 1500 data points were collected. For straight and curved segments of highways， the occupant injury evaluation models of trucks and cars were fitted based on the ASI. Third， according to the Fisher optimal segmentation method， reasonable classification standards of risks of roadside accidents and the corresponding ASI thresholds were determined， and the risk assessment methods for roadside accidents based on the ASI were provided and verified. Finally. a proportion of trucks was introduced to further improve the ASI evaluation model. The results show that the ASI has a positive linear correlation with the departure speed and subgrade height， an exponential correlation with the side slope， and a power correlation with horizontal radii. Setting the roadside guardrail can reduce the roadside accident injuries of cars by 24%~28% and that of trucks by 31%~36%. Compared with cars， trucks are more prone to serious roadside accidents.

In order to recognize road traffic state timely and effectively， a method combining auto-encoder with classifiers was proposed for traffic state recognition from videos. Firstly， traffic video and image data sets were built， based on which auto-encoder structure parameter tests and optimizations were conducted over hidden layer and reduced data dimensionality. Then， a quantitative evaluation method for auto-encoders was put forward， thus the best auto-encoder model was presented as A^*. At last， four traffic state recognition models were constructed by combining A^* with Linear Classifier （LC）， Support Vector Machine （SVM）， Deep Neural Network （DNN） and DNN-LC respectively. The before mentioned models and CNN models including AlexNet， LeNet， GoogLeNet and VGG16 were trained and tested. Results show that， precision and recall of the proposed models are 94.5%~97.1%， and their F1 values are 94.4%~97.1%. Furthermore， AlexNet performs best among the four CNN models， with precision， recall and F1 value equal to 94%. Therefore， combiningA^* and commonly used classifiers may reach or surpass the traffic state recognition effects of complicated CNN models. The proposed methods are convenient for training and testing with low computation cost， which are suitable for traffic state recognition from videos or images.

In order to reduce the carbon emissions based on the premise of specified transportation tasks in the large manufacturing enterprises， first， the account of the feasibility and superiority of adopting drop and pull transport is given out. Second， an optimization dispatching model of tractors is constructed under the constraint of hard time window， with t·km CO₂ emission as the objective function. Third， a two-stage algorithm is designed for solving the optimization dispatching model. The algorithm presented applies scan algorithm to get the initial feasible solution first， then adopts the simulated annealing algorithm and tabu search algorithm respectively to improve the solution. The optimization model and the algorithm proposed are applied to the 11 data sets of Hunan Hualing Iron and Steel Company. The experimental results indicate that the optimization dispatching model and the two-stage algorithm are feasible and effective. The dispatching optimization method for tractors with carbon emission consideration proposed in this paper has a good effect on energy saving and emission reduction. It also indicates that the reasonable location of the center node of tractors in large manufacturing enterprises can effectively decrease the carbon emissions of drop and pull transportation.

To solve the problem of workers' load imbalance in the flow shop scheduling， a dual-objective optimization scheduling model is proposed in this paper with the minimum delay time and the workers' workload standard deviation. A NSGA-II based on two-gene chromosome coding is designed to obtain Pareto-optimal solutions. Two embedded heuristic rules， the earliest due date （EDD） rule and the shortest processing time （SPT） rule， are introduced together with NSGA-II to form the NSGA-II-EDD and NSGA-II-SPT for comparison. Computation experimental analysis shows that NSGA-II performs better in case of evaluation indexes with the average non-dominated solutions N， error ratio ER，spacing evaluation index S and Pareto front span K， but is worse in operation time T.

In order to solve the Just-in-Time （JIT） material distribution scheduling problem for mixed-model assembly lines， a static semi-kitting strategy is presented. First， the material distribution scheduling problem based on the static semi-kitting strategy is described and a multi-objective optimization model is established to minimize the Total Energy Cost （TEC） and the Total Line-side Inventory （TLI）. Then， an improved multi-objective gravitational search algorithm is proposed. The chaotic gravitation operator and the memory search strategy are introduced to the proposed algorithm to accelerate the convergence speed and increase the population diversity. In addition， a local search optimization operator is constructed to optimize the TEC and the TLI. Finally， the experimental results prove the feasibility and effectiveness of the proposed algorithm by comparing with benchmark algorithms.

The combined prefabricated box culvert with four members is a new type of culvert structure. Based on the construction project of the new expressway between Songyuan and Tongyu section of Tieke expressway in Jilin Province， the field test of the new culvert structure is carried out， and the stress and deformation characteristics of the structure during the construction of different layers are analyzed. The finite element model of culvert-soil structure is established. The theoretical and measured results are compared to verify the correctness of the model. The results show that with the continuous increase of the layered filling height， the upper roof continuously recedes and the left and right side walls continuously protrude， and the selection of backfill soil with large elastic modulus in the backfilling process has a positive impact on the stress and deformation performance of the box culvert. The influence of vehicle live load on bearing deformation of culvert decreases with the increase in the filling height.

To analyze the influence of the content of aggregate on the mechanical properties of asphalt mixture， four grades of asphalt mixtures are prepared by Superpave gyratory compactor. The cracking resistance， compression resistance， and creep properties of the four asphalt mixtures are respectively studied by splitting tests， compression tests， and creep tests at different temperatures. Then， the variations of relaxation moduli of four grades of asphalt mixtures with the loading time are deduced by convolution integral and Laplace transform， and the creep and relaxation characteristics of four grades of asphalt mixtures are fitted and analyzed by Burgers model and second-order extensive Maxwell model. Finally， according to the grey correlation degree algorithm， the correlation degree between aggregate content and mechanical properties of asphalt mixtures is calculated， and the influence of aggregate content on the mechanical properties of asphalt mixtures is analyzed. The results show that the low-temperature cracking resistance， room temperature tensile and compressive properties of asphalt mixtures are highly related to the content of aggregate with 0.15~0.3 mm and 1.18~2.36 mm， while the creep resistance at room temperature is mainly affected by aggregate with 0.6~1.18 mm and 2.36~4.75 mm. As the temperature increases， the effect of aggregate content on the creep resistance increases. The relaxation strength is greatly affected by the fine aggregate， while the relaxation time is mainly affected by the larger aggregate.

Coupling agent KH-550 and hydrochloric acid are used for basalt fiber surface treatment to investigate the effect of modification conditions on the mechanical properties and working performance indexes of basalt fiber-reactive powder concrete. The optimum modification conditions were selected as follows： 0.75wt% of coupling agent， 3 mol/L of hydrochloric acid， acid etching temperature of 20 ℃ and etching time of 60 min. By using acoustic emission （AE） technology， the damage characteristics of basalt fiber reactive powder concrete before and after modification were compared to determine the fracture stage and fracture mode of the sample in the fracture test. The results show that in AE parameters， the accumulative AE hit， accumulative energy and amplitude are related to the damage stage of basalt fiber reactive powder concrete， Fiber modification affects the final stage of concrete loading. In addition， during the loading process， the rise angle （RA） and the average frequency （AF） have opposite trends， and their changes are related to the fracture mode.

The mechanical characteristics of the steel-plypropylene hybrid fiber concrete （SPFRC） shear walls were studied by cyclic testing on two SPFRC shear walls and one reinforced concrete （RC） shear wall under reversed cyclic loading．The effects of fiber content and axial load ratio on the failure pattern， shear capacity， ductility and energy dissipation capacity of the shear walls are analyzed. The test results show that under the same axial load ratios， the hybrid fiber can effectively restrain the development of shear wall cracks， improve the shear capacity， deformation capacity and energy dissipation capacity of the shear wall. For hybrid fiber specimens， with the increase of axial load ratio from 0.1 to 0.2， the bearing capacity and ductility are improved， but the energy dissipation capacity declined. According to the mechanism of truss-slope lever mechanism， the SPFRC shear wall shear bearing capacity calculation equation was established， in which the contribution of horizontal and vertical distribution of rebar， concrete oblique lever and concealed columns to the shear-bearing capacity are taken into consideration. This equation is verified using domestic relevant data. The average value of the ratio between measured and calculated shear bearing capacity is 1.01 and the standard deviation is 0.17， which are well matched.

This paper aims to investigate the effect of void characteristics on the virtual shear fatigue life of asphalt mixtures. Using image techniques and three-dimensional （3D） Discrete Element Method （DEM）， the micromechanical models for the shear fatigue life of asphalt mixtures were established considering the real void size and distribution and random void size and distribution. The shear fatigue lives of asphalt mixtures with different aggregate sizes， binder contents， and temperatures were simulated based on these models. Some simulation results were compared with the experimental results. Research results show that the shear fatigue life of asphalt mixtures can be simulated using 3D DEM， and void characteristics significantly affect the virtual shear fatigue life. With the same model parameters， the shear fatigue life considering the real void size and distribution is less than that considering the random void size and distribution， but is closer to the experimental results. Furthermore， aggregate size and binder content also remarkably affect the shear fatigue life of asphalt mixtures. The shear fatigue life increases with the nominal maximum aggregate size （NMAS）. Under the same conditions of NMAS and temperature， the shear fatigue life is the lonest with the optimal binder content. Under the same conditions of NMAS and binder content， the shear fatigue life decreases as the temperature increases.

In order to meet the requirement of efficient and quantitative concrete quality defect detection， a fast imaging algorithm based on the integral equation method with Biconjugate Gradient Stabilized-Fast Fourier Transform （BCGS-FFT） and the Multiple-source and Multiple-frequency Contrast Source Inversion （MMCSI） is proposed. Firstly， the FFT is introduced into the forward process to accelerate the calculation efficiency of the integral equation green's function， and the BCGS is used to improve the stability of iterative solution in the large matrix equations. Secondly， the CSI algorithm without calculating Jacobian matrix is introduced in the inversion process， which can be applied to the data volume of special observation system. The FFT is applied to speed up the matrix multiplication calculation involved in the iterative process of inversion. Finally， numerical simulation and practical examples show that the algorithm can obtain better reconstruction results and improve the efficiency of location in concrete quality defects detection. Furthermore， the results verify the feasibility and effectiveness of the algorithm， which lays a theoretical foundation for the application of acoustic detection to quickly extract the properties of concrete quality defects.

In this paper， experimental and nonlinear finite element analysis methods are used to study the mechanical characteristics of the joint area. It is clear that the shear bearing capacity of the joint area is mainly borne by the components of the joint area， and the contribution of the components of the joint area to the shear bearing capacity is analyzed；At the same time， through the analysis of the stress mechanism of each component in the joint area， the calculation method of shear capacity of each component in the joint area is derived；Finally， the shear capacity formula of the new joint is obtained by superposition principle，Compared with the test and finite element analysis results， the formula is safe and reasonable， and can meet the engineering seismic design principle of “strong joints and weak members”.

This paper proposes a Chinese named entity recognition method based on Transformer encoder and BiLSTM. This method uses a joint vector as the word representation layer by combining the word embedding and the position coding vector to avoid the losses of the word embedding information and the position information. The directional information is integrated into the joint vector using BiLSTM. The Transformer encoder is introduced to further extract the word relationship features. The experimental results show that the F value of this method on the general MSRA and Thangka domain data sets reaches 81.39% and 86.99% respectively， which effectively improve the effect of Chinese named entity recognition.

In order to overcome the shortcomings of traditional infrared and visible image fusion methods，a fusion method based on Discrete Nonseparable Shearlet Transform （DNST） and Convolution Sparse Representation （CSR） is proposed. Firstly， the source images are decomposed into approximate images and directional detail images using DNST. Compared with other multi-scale decomposition tools， DNST can better separate the overlapped important feature information in different scales. Secondly， the salient feature maps of the source images are employed to weight average approximate images， which can prevent the loss of the brightness and energy. CSR can deeply extract the salient features of the image， The l1 norm of multi-dimensional coefficients is used as the activity level measure to construct the Salient Feature Map （SFM）， which can generate the weight distribution decision map of the approximate image. The rule of Coefficient absolute max-Gaussian filtering is used as fusion rule of the directional detail images. The decision map of initial weight distribution is obtained by the Coefficient absolute max rule， then the decision map is filtered by Gaussian filter to reduce the sensitivity of noise and increase the proportions of visible image information. Finally， the fused coefficients are reconstructed by the inverse DNST， and the fusion image is obtained. Experimental results demonstrate that the proposed fusion method can achieve superior performance compared with other typical fusion methods in the existing literature in both subjective vision and objective criteria evaluation.

In order to improve the accuracy and recall of geospatial data integration extraction， a geospatial data extraction algorithm based on machine learning is proposed. GeoNames， OpenStreetMap， etc. are used as the data sources of geographic information. Through web crawler and search engine， the relevant web pages are searched and downloaded at the same time， and the content is filtered. After filtering， the location name and address information and other data in the web pages are parsed and extracted to realize visualization. By analyzing and extracting the geographic data entities， using the mapping between geographic data and entities， the disambiguation of heterogeneous geographic data is eliminated， the integration of geospatial data is realized， and the digital features of geographic data are realized according to the similarity degree calculation of multi features such as entity name and category. Combined with multi feature and machine learning KNN classification method， the proposed algorithm can complete the automation of geographic data link and realize the classification and extraction of geospatial data. The experimental results show that the proposed algorithm has high precision and recall， and the data extraction effect is good， which can lay a foundation for the integrated extraction of geographic data.

In this paper， having deeply studied the fault output characteristics in circuits and the ADD candidate diagnosis solution method， we propose a candidate diagnosis solution method under the guidance of fault response， ALFDD， namely. We also put forward the concept of candidate single fault set， which is based on the principle that a single fault with different actual output response and expected output response under current test excitations is more likely to be a fault diagnosis solution. To overcome the problem that in ADD method to obtain Fsame all single faults of circuit should be compared， we give the method which only considers the faults of the candidate single fault set to obtain Fsame. This method has three merits. Firstly， it does not need to compare all single faults and improves the efficiency of solving problems. Secondly， it eliminates the reluctant candidate solutions concluded in Fsame effectively， so it improves the resolution of candidate diagnostic solutions. In addition， this method improves the accuracy of candidate diagnostic solutions， namely， increasing the number of real diagnostic solutions in the candidate fault sets. The test results show that，， compared with ADD， the proposed ALFDD method increases the resolution， accuracy and solution efficiency for selecting candidate diagnoses.

When the traditional detection methods are used to identify User Interface （UI） components， there is no way to categorize these components. To solve this problem in order to provide guidance for designers to reconstruct complex UI screenshot examples， this paper proposes an improved method based on improved YOLOv3 for UI components detection. First， the feature extraction network uses tightly connected network （DenseNet） to make full use of the extracted features. Second， channel attention mechanism and spatial attention mechanism are added to the dense layers and transition layers of the feature extraction network， the weighted feature is used to replace the original feature for the later feature fusion. Third， the feature pyramid network is constructed to complete the components detection task in four dimensions. Finally， the focalloss is used as the classification loss function. The data set consists of a large number of real Android application interface screenshots and XML files. The experimental results on the collected data set show that the recall rate of the method is 91.97% and the mAP is 48.21%， which has better performance than the traditional detection methods.

In order to obtain the ideal network traffic prediction results， a large-scale network traffic prediction model based on cloud computing and big data analysis is proposed. Firstly， according to the chaos algorithm to describe chaos characteristics of network traffic， the learning sample set is established. Then the support vector machine is introduced to model the randomness characteristics of network traffic， and combined with the massive characteristics of historical data， the cloud computing platform is used to make multiple support vector machines run in parallel. Finally， the comparative test results show that the proposed model improves the accuracy of network traffic prediction， and the modeling efficiency is greatly improved， which can meet the real-time requirements of online network traffic management.

The Equal-Cost Multi-Path （ECMP） algorithm does not consider the characteristics of network load and traffic flow， that it is easy to map multiple big data streams to the same path， resulting in network bottleneck link. To overcome this shortcoming， a dynamic traffic scheduling （DTSNL） scheme based on network load for Software Defined Network （SDN） data center is proposed. The scheme combines the characteristics of network load and traffic to schedule the traffic reasonably and achieve network load balancing. The controller calculates the traffic threshold by periodically counting the traffic information of the access layer switch in the Fat-Tree network topology， and chooses the best path for the large data flow with high bandwidth ratio. The simulation results show that， compared with the ECMP scheme and the GFF scheme， the poposed DTSNL scheme improves the network average throughput， link utilization， and core switch load and link bandwidth utilization.

The overall structure information is obtained in the training of the capsule graph neural network, and as the layers increases, the structure feature information of the node will be lost. A capsule graph neural network that combines global and local features was proposed. First， the Node2vec is improved， and the attribute information of nodes is introduced into the random walk process， so that the network structure and the attributes of nodes are taken into account when the network representation is generated. Then， the improved Node2vec is introduced into the capsule graph neural network， and the capsule graph neural network is designed which fuses global and local characteristics. Experimental results show that the proposed capsule graph neural network has faster training convergence， and higher graph classification accuracy.

Because the existing methods fail to consider the problem of poor anti difference， the quality of image encryption is reduced and the security performance is poor. In order to effectively solve the above problems， combined with the blockchain technology， a dual chaotic recognizable tamper image encryption method based on blockchain technology is designed and proposed. Kawakami hyperchaos are used to form hyperchaotic sequences for image pixel replacement. Combined with blockchain technology， the initial value of 3D Bao system is formed. The encryption sequence is formed by using plaintext pixel value and Bao system. The encryption sequence is replaced and diffused by the encryption sequence， which effectively realizes the encryption of double chaotic recognizable tamper image. Simulation results show that the proposed method can effectively improve the quality of image encryption and enhance the encryption effect.

To overcome the problem that the effect of optical image encryption is limited by the processing technology of the optical encryption devices and the manufacturing process of the random phase mask is complicated， in this paper， an optical image encryption learning scheme based on cycle-consistent adversarial networks （CycleGAN） is proposed. Firstly， the classic double random phase encoding is used to encrypt the plain image to generate a plain-cipher training set. Secondly， the training set is input to CycleGAN to automatically learn the encryption characteristics of optical image encryption to obtain an optical image encryption learning model. Finally， encryption and decryption performance tests are carried out by simulation experiments on the images generated by the learning encryption mechanism of CycleGAN. Data analysis shows that this scheme can effectively protect the security of image information and recover ciphertext images well. In addition， the encryption performance is not limited by optical encryption equipment， which can realize the rapid encryption of batches of images.

In the current research of abnormal behaviors in underground mines the result accuracy needs improvement. In order to solve this problem and improve the safety of underground mines， a continuous density hidden Markov model is proposed to identify abnormal behaviors in underground mines. In the detection of human motion area under the mine， first， the video frame data to be identified is obtained； the human motion area is preliminarily extracted for each image， and the preliminary detection of human motion area is realized by cascading classifiers to get more accurate human motion area. Then the next frame data is read in， and this process is iterated until all frames are detected. According to the preliminary detection results， the continuous density hidden Markov model is introduced to decompose the human body image into several equal areas. The representative color features and the standard difference features in the image area are obtained. The continuous density HMM of the target is trained through the obtained feature data. The abnormal behavior recognition of the human body target under the mine is completed according to the training model. The experimental results show that the proposed algorithm has the characteristics of high accuracy and high detection rate in different number of human behavior recognition， which shows that the algorithm is reliable.

In view of the complex structure of the current commonly used filtering algorithms， low work efficiency and single Digital Elevation Model （DEM） results evaluation methods， this paper proposes a moving surface fitting DEM filtering algorithm based on spatial grid technology and convolution kernel calculation （CGS） and a new method of DEM quality evaluation based on computer graphics. The results of theoretical research and engineering practice show that the CGS algorithm can better filter out various non-ground points such as vegetation and buildings than the manual intervention of the TIN algorithm. The gray-level co-occurrence matrix， autocorrelation function， image signal-noise ratio， The improved local variance algorithm of Gaussian waveform extraction based on Canny operator are used to compare and analyze the results of DEM. It is found that the DEM image generated by the CGS algorithm has a uniform and smooth texture， which highlights the detailed features of various terrains and can effectively suppress the generation of salt-and-pepper noise. The filtering algorithm and evaluation system proposed in this paper are forward-looking in theory， scientific and repeatable in method， which can be applied to actual project engineering and have strong practical significance.

In order to solve the problem of path planning and tracking control for high-speed lane changing collision avoidance of vehicle， a path planning method and a path tracking control strategy for high-speed lane changing collision avoidance of vehicle are proposed based on quintic polynomial curve and discrete sliding mode control theory， respectively. Firstly， a feasible path is planned based on quintic polynomial curve. The mapping relationship between the maximum curvature， the maximum curvature change rate of the planned path and the undetermined coefficients of the quintic polynomial curve is established indirectly by lookup table， so that the planned path meets the requirements of riding comfort and safety. Secondly， in order to track the planned path based on quintic polynomial curve quickly and steadily， a linear discrete control model with additive uncertainty is established by combining the vehicle kinematics model with the linear vehicle dynamics model of two degrees of freedom， and a path tracking control strategy is designed based on discrete sliding mode control theory with disturbance observer. Finally， a model in the loop simulation system is established based on the software of vehicle dynamics. The feasibility and effectiveness of the proposed path planning method and path tracking control strategy are verified by the model in the loop simulation system.

Based on the single periodic potential system， a new method of stochastic resonance signal enhancement for coupled periodic potential systems was proposed. The method uses particle swarm optimization to achieve adaptive matching of system parameters， coupling coefficients and step sizes. Using this method the targeting signal and the signal-noice-ratio can be enhanced. Experiments are carried out to verify the ne method. The results show that due to the function of the coupling system， the control system affects the stochastic resonance of the controlled system by adjusting the parameters， so that the adaptive coupling periodic potential system stochastic resonance method performs better than the adaptive first-order periodic potential system stochastic resonance method in the enhancement of weak fault characteristic signal. Due to the synergy between the stochastic resonance of the control system and the stochastic resonance of the controlled system， the stochastic resonance effect of the controlled system is greatly enhanced. Therefore， the adaptive double-input coupling periodic potential system stochastic resonance method is more suitable for the extraction of weak fault characteristic signals in the noise environment than the adaptive single-input coupled periodic potential system stochastic resonance method.

State constraints， input constraints and external disturbances usually exist in the path following problem of wheeled mobile robots. Based on nonlinear disturbance observer， a moving horizon control strategy for path following problem of wheeled mobile robots is proposed in this paper. While there is no disturbance at all， the moving horizon control can satisfy the input and state constraints， and drive the wheeled mobile robot to the desired path. While there are disturbances， in particular， slow varying and “big” disturbances， the proposed nonlinear disturbance observer can estimate the disturbances， and compensate the influence of the disturbances on the wheeled mobile robot through a feedback. Simulation results show that the proposed control strategy can guarantee the convergence of the mobile robot to the desired path under the external disturbance.

When the conventional method is used to predict the constant flow rate of the variable speed hydraulic power source， the time used to predict the flow rate is long， the error between the predicted result and the real rate is large， and the prediction accuracy is poor. To overcome the above drawbacks， based on the deep learning and limited amplitude fuzzy， a constant flow prediction method of variable speed hydraulic power source is proposed. Using the mathematical models of permanent magnet synchronous motor， servo controller， gear pump and proportional relief valve， with the principle of deep learning， the constant flow of variable speed hydraulic power source is predicted by the limited amplitude fuzzy control technique. The experimental results show that the proposed method has higher prediction efficiency and accuracy.

In order to solve the problem that particle filter tracking algorithm uses single feature with lower robustness and the resample strategy is easy to cause particle degradation and impoverishment， an improved particle filter tracking method based on multi-feature and multi-cue is proposed. First， the Histon histogram that introduces the neighborhood relationship is used to describe the color characteristics of the target. Then a robust fuzzy statistical texture feature is used to express spatial texture information， and it is adaptively fused with regional color features to construct a multi-cue observation model. In the particle filter tracking process， K-means-based particle weight clustering is used for more accurate posterior distribution estimation. In the importance resample stage， high-weight particles are retained while new particles are generated based on the prior distribution of the current target state， which ensures particle diversity and avoids particle degradation. Experiments are carried out on the standard test set. Compared with other tracking algorithms based on particle filter framework， the proposed method obtains higher tracking accuracy and success rate. Compared with other popular tracking algorithms， the proposed method can achieve better tracking results under illumination variation， target deformation and background disturbance scenarios.

The multiple frictional contact dynamics of the flexible threshing tooth against grains were presented using an addition-deletion constraints approach. During threshing， the grains undergo impacts and rubbing as they fall from straw to become free particles， thus， they experience a serious of frictional collisions and mutual sliding with tooth. An accurate model of a flexible threshing tooth was established on the basis of high-order rigid-flexible coupling theory. The threshing process was characterized by load-off， multiple sliding-stick collisions and separation. An addition-deletion constraints approach was used to solve the dynamic equations for separation， initial contact， stick， and slip. The contact detection and prediction were made with every integral step. Once contact-pairs between threshing tooth and grains were detected through searching algorithm， the multiple contact constrains as well as constrains equations were activated. Corresponding computational strategy software in the C programming language was developed with explicit solution method. It can obtain the vibration and deformation of threshing teeth， as well as the normal impact force and the tangential stick-slip kneading force of threshing teeth against grains. EDEM software simulation and experimental verification were carried out on the longitudinal flow flexible threshing drum model. The results show that the flexible tooth exerts a less impactful kneading force than the rigid pole tooth on cracked grains or seed， thereby reducing the rate of grain damage， and is thus favorable for increasing the combined benefits to paddy production.

Due to the complexity of the insect environment in the field and the imbalance in the number of samples among insect categories， the existing automatic identification and classification methods for field insects have high misidentification rates and low efficiency. In this paper， a new field insect automatic identification and classification algorithm is developed based on a lightweight deep learning model. First， preprocessing applied to the picture， then those images were input to the lightweight algorithm for feature extraction， and multi-scale feature fusion were adopted to output prediction networks of different sizes； then introduce joint cross-comparison for automatic identification and classification of field insects， and finally compare with the reference The algorithm has been simulated and compared. The results show that the field insect automatic identification and classification of the algorithm in this paper has high accuracy， less time， and strong robustness. It effectively solves the problems of insect accumulation and background interference and can identify field insects in real-time and online.

In order to meet the requirements of structural safety， in this paper， first， the characteristics of high strength ratio and high mechanical properties of honeycomb structure were used for reference， and applies the principle of structural bionics to design and establish three simplified structure models， the direct honeycomb structure， oblique honeycomb structure and plate structure. Second， under five different working conditions， the finite element analysis software OptiStruct was used to analyze the bearing characteristics of three models. Finally， the three models were prepared by 3D printing technology， and the mechanical properties of the samples were tested and compared with that of simulation results. It is shown that the stress values at each position of the structure are similar to each other， and the structure has good mechanical conductivity. Under the condition of the same quality， the bearing capacity of the imitation oblique honeycomb structure is 12% higher than that of the imitation vertical honeycomb structure， and 150% higher than that of the plate structure. This work may provide a reference for the lightweight design of the protection structure.

To ensure the safety of the sampling task on lunar surface， the lunar sampling arm should contact with lunar soil in advance to identify the lunar soil compactness. Based on geometric parameters of the indentation between the manipulator and lunar soil obtained by the stereo cameras， a model was established to identify mechanical performance of the lunar soil using least squares support vector machine. A total of 264 data were collected. The data was split into two data set randomly， one was calibration set （contains 96 data） and the other was prediction set （contains 48 data）. Indentation length， deep， area and cube were used as input parameters to establish the prediction model， and the accuracy of prediction set are 93.75% of CE5_1， 83.33% of CE5_2， and 87.50% of CE5_3， respectively. The results show that this prediction model can be used to identify mechanical performance quickly which can be used as a method to determine the sample’s depth for lunar soil surface.