In order to solve the problem of single-object grasping detection of robotic arms， a SqueezeNet model algorithm based on attention mechanism （CBAM） was proposed. Firstly， the deep vision grasping system was described， and the hand-eye calibration of the grasping system was completed. The data set was preprocessed by randomly clipping， flipping， adjusting contrast， and increasing noise， effectively expanding the object capture data set. Secondly， a lightweight SqueezeNet model was introduced. It uses a five-parameter method to characterize the two-dimensional grab frame， which can complete the target capture without increasing the difficulty of network design. Thirdly， a plug-and-play network with an attention mechanism was introduced to weight the incoming feature maps in the channel and spatial dimensions. The SqueezeNet model-grabbing network was optimized and improved. Finally， the improved CBAM-SqueezeNet algorithm was verified on the public data sets Cornell grasping dataset and Jacquard dataset. The grab detection accuracy is 94.8% and 96.4%， accuracy increased 2% than the SqueezeNet network. The CBAM-SqueezeNet network grabbing method has a reasoning speed of 15 ms， which balances grabbing accuracy and running speed. The paper conducted experiments on the Kinova and SIASUN arm， and the success rate of network capture was 93%， which was faster and more efficient.

The subspace pursuit-polynomial chaos expansion method （SP-PCE） was proposed for the uncertainty quantification of EV-WPT system's transfer efficiency. Firstly，by establishing the three-dimensional electromagnetic simulation model of EV-WPT system and reasonably setting the distribution type of relevant variables， the statistical characteristic parameters such as mean， variance and probability density curve that can characterize the uncertainty of EV-WPT system's transfer efficiency were calculated by SP-PCE method. Then， combining the SP-PCE method and Sobol method to carry out the global sensitivity analysis to obtain the influence degree's quantitative index of random input variables on system's transfer efficiency. Finally， the accuracy and efficiency of the proposed method were verified by numerical simulation experiments， which provides a theoretical basis for ensuring the high transfer efficiency of EV-WPT system as well as system's structure optimization.

In order to improve the comprehensive work efficiency and energy supply stability of the automotive composite power supply， a power demand control method of the idle start stop hybrid electric vehicle composite power supply was proposed. The control requirements of the composite power supply was analyzed through its composition， and the states of batteries and supercapacitors were clarified. According to the dynamic calculation of the relationship between the maximum speed of a hybrid vehicle and the supercapacitor， the input state compensation amount and the actual supercapacitor were used to obtain the maximum limited output power. By using the recursive method of multi-scale decomposition of discrete sequences， the scale space was gradually decomposed， and the decomposition and reconstruction coefficients were calculated to achieve the power demand control of hybrid electric vehicle composite power supply. The experimental results show that after using the proposed method for control， the capacitance curve remains at around 0.60 F， the current fluctuation amplitude is -50~50 A， and the power fluctuation amplitude is relatively gentle， with a voltage between 200~220 V. This indicates that the proposed method can stably control the power demand of the idle start stop hybrid electric vehicle composite power supply， reduce the energy consumption of the entire vehicle， and ensure maximum resource utilization.

To increase the operating range and improve the flexibility of the mechanism， a scissor aerial work platform with a rotating and telescopic power device was developed. Based on the preliminary design of the basic dimensions of the scissor aerial work platform， an analysis of the anti-overturning stability was conducted. When the operating platform was rotated to an angle of 90° with the bottom plate， the mechanical model was established to verify its load stability requirements by the stability coefficient method， and the speed and acceleration of the working platform were calculated using the cartesian coordinate method and the instantaneous velocity center method. The hydraulic cylinder thrust was obtained by the principle of virtual work， and the hinge point position of the lower hydraulic cylinder was optimized by genetic algorithm. Finally， dynamic simulation and comparative analysis were performed using ADAMS， and combined with the physical acceleration and velocity measurement results of the work platform， the correctness of the theoretical calculation and simulation results is verified.

Hot compression test for 2.25Cr-lMo-0.25V steel was carried out at the temperature of 950-1 200 ℃ and strain rate of 0.005-0.1 s^-1 on Gleeble-1500D thermal simulator. Hot deformation behavior and dynamic recrystallization of this steel were systematically investigated. The flow stress and microstructure state were strongly depended on temperature and strain rate. The relationship between Zener-Hollomon parameter and the characteristic strain （critical strain and steady strain） was determined by linear fitting， and a dynamical recrystallization state diagram was established. The dynamical recrystallization kinematic equation of this steel and the model of complete dynamic recrystallization grain size were established based on experimental data and the stress-strain curves. The results could provide a theoretical basis for formulating reasonable hot working process.

Experimental and simulation results were combined with neural network and genetic algorithm methods to study the non-uniform mechanical properties of DP980 laser welded joints. The shear correction Gurson-Tvergaard-Needleman （GTN） model was obtained to optimize the damage parameters of the welded joints. The optimization results were substituted into the finite element model and compared with the experimental results to verify the accuracy of the fitting parameters， and analyze the cause of transverse crack propagation in cup-shaped specimens through changes in pore volume fraction. Accurate mechanical properties of welded joints were obtained through the above methods to analyze and predict fracture damage of welded joints under complex stress states.

Using Abaqus finite element software， 150 finite element models of oblique T-shaped welded joints with different angles， plate thickness ratios， and length-to-thickness ratios were established to analyze and obtain the maximum hot spot stress concentration factors （HSCFs） at critical points. Based on the Levenberg-Marquardt algorithm combined with a general global optimization method， the data were fitted to derive simplified and refined formulas for the maximum HSCF. The results indicate that the HSCF at the toe of the obtuse angle side is always greater than that at the acute angle side， and as the angle difference decreases， the difference in HSCFs between the two sides correspondingly diminishes. The HSCF at the toe of the oblique T-shaped joint shows a positive correlation with the angle， plate thickness ratio， and length-to-thickness ratio， reaching its maximum when the angle is 80°， the plate thickness ratio is 2， and the length-to-thickness ratio is 18. The proposed formulas for calculating the maximum HSCF show good agreement with the finite element numerical results， with the correlation coefficient R2 of the refined formula reaching 99.97%.

The fatigue crack propagation behavior of 2024 aluminum alloy used in aerospace was studied under air and corrosion environment， and the effect of ultrasonic rolling on the fatigue crack propagation behavior was analyzed and compared. The results show that the surface grain of the material is refined by ultrasonic rolling treatment to form a plastic deformation layer with a thickness of 60-70 μm， and the microhardness is increased by 45% compared with the base metal， and the gradient decreases from the surface to the interior of the material. The initial rate of fatigue crack growth was decreased and the growth rate of fatigue crack growth was increased by ultrasonic rolling treatment. Ultrasonic rolling can significantly improve the fatigue performance in both environments， but the improvement is more significant in the air environment， because the saline environment increases the initial crack growth rate and reduces the growth rate of crack growth rate.

The specific influencing factors of the arrival status of high-speed trains on railway passengers' choice behavior were studied. First， an evolutionary game model of the arrival status of high-speed trains and railway passengers' choice behavior was constructed. Then， the evolutionary behaviors of both players in the game were analyzed， and the evolutionary equilibrium strategies of the replicator dynamic system were obtained. Finally， simulation verification was carried out， and the results indicated that there are two evolutionary stable states. When the delay time of the high-speed train is within 30 minutes， the train tends to adopt a punctual arrival strategy， and passengers tend to choose to travel. When the delay time exceeds 30 minutes， the train tends to adopt a punctual arrival strategy， while passengers tend to choose to refund their tickets. The game parameters can be adjusted to direct the evolutionary direction of both players， which can effectively improve the proportion of railway passengers choosing to travel when the high-speed train is delayed.

A method for detecting drivers' angry driving behavior has been designed using widely used popular smart bracelet， which provides a new way and method for effective monitoring of angry driving behavior. 50 drivers were recruited to conduct a simulated driving experiment， and a simulated driving scene that caused anger was designed. Then， heart rate index HR and eight heart rate variability （HRV） indexes such as RR.mean， SDNN， RMSSD， PNN50， SDSD， HF， LF and LF/HF obtained from bracelet collection data were used to study the correlation between the acquisition indexes and the angry driving behavior， and screen the significant influence indexes Finally， using three methods， namely support vector machine （SVM）， K-nearest neighbor （KNN） and linear discriminant analysis （LDA）， established and verified the detection model of angry driving behavior. The results show that the model based on KNN algorithm has the best performance on anger recognition. The accuracy of anger intensity recognition can reach 75%， and the accuracy of anger state recognition is 86 %. The results show that the wearable device （smart bracelet） can reasonably detect the driver 's anger state and anger intensity.

Cold-formed thin-walled steel and glued wood were combined into a new type of box-section beam through different connection methods. Taking the steel and wood connection methods and steel crimp size as variables， three-point bending tests were carried out on four combined beams to investigate the damage mechanism and mechanical properties of the combined beams. The test results show that the overall performance of the combined beams is good， the deformation is coordinated， and the combined beams are finally destroyed by the fracture of the glued timber at the lower flange of the pure bending section. Under the different connection modes， the ultimate bending capacity of the glued-connected combined beams were improved by 61.5% and 23.0% compared with the bolted-connected and tapping screw-connected ones， respectively， in the same way as the glued-connected combined beams， the bending capacity of the combined beams is only improved by 9.0% by enlarging the dimensions of the steel crimps. A finite element model of the combined beam was established based on the test， and the simulation results were in good agreement with the test results. The finite element simulation results showed that the steel strength， steel thickness， and thickness of glued wood all had a certain effect on the bending capacity of the combined beam， with the thickness of glued wood having the greatest effect. Combined with the test results and simulation analysis of the cold-formed thin-walled steel-glued laminated timber combination beam span deflection and flexural capacity calculation formula， the theoretical values and simulation values are in good agreement， which can provide a certain reference for engineering practice.

The issue of distinguishing driving behavior characteristics in steep sharp curves was addressed by employing a behavior classification method based on vehicle driving parameters. After analyzing real driving data， six variables were selected， including vehicle type， velocity before curve， velocity after curve， deceleration before curve， car-following driving and differences in velocity pre- and post-curve. The characteristics of each type of drives were determined based on the construction of the latent model. The findings that drivers could be classified into three categories： stable drivers， restricted drivers， and free drivers. Stable drivers exhibited parameter values that fell between all three types， they tended to enter and exit curves at a consistent speed. Restricted drivers showed significant features of deceleration or car-following driving before curves. Conversely， the free drivers had the highest velocity before and after curve with their differences in velocity pre- and post-curve were also the highest.

In order to account for the variations in construction progress， the material parameters， connection stiffness， and boundary conditions of the deep foundation pit are modified. This results in the utilization of traditional deterministic finite element modeling parameter method being inefficient， slow， and difficult to converge. The following problem is proposed to address this issue： The probabilistic finite element parameter correction of the baseline model concept involves the establishment of a characterization of the subway deep foundation pit in the different stages of the construction of physical parameters. In essence， a finite element model is established to characterize the changing law of physical parameters in different construction stages of subway deep foundation pit. Initially， the Gaussian mixture model is employed to partition the maximum horizontal displacement of the enclosure wall into multiple clusters， thereby extracting the mean value and variance. Subsequently， the ANSYS finite element deep foundation pit model is configured， and finite element corrections are implemented on the parameters of internal support and enclosure wall of the deep foundation pit. Finally， the probabilistic baseline of the deep foundation pit finite element model for cluster analysis is obtained by using the Kriging prediction model instead of the analytical finite element. The results show that the computational rate and accuracy of the baseline model are significantly improved after the parameter correction， which can provide a theoretical basis for the subsequent numerical simulation.

In order to ensure the safety of the building， a method for monitoring the seismic resistance of the assembly joints of corroded reinforced concrete column under heavy rainfall was proposed. Firstly， the reasons for the corrosion of reinforced concrete columns caused by heavy rainfall environment were analyzed， and the deterioration degree of reinforced concrete columns caused by rain rust was calculated.Then， in order to clarify the influence of rain rust on the seismic resistance of reinforced concrete columns， four reinforced concrete specimens with different degrees of corrosion were constructed with reference to the design standards for reinforced concrete structures in the construction industry. The four specimens were placed in a pseudo-static experimental environment， and different vibration environments were simulated by loading different degrees of loads. Finally， by combining the calculation of hysteresis curves， skeleton curves， and inter story displacement angles， the seismic performance analysis of different specimens under the same vibration environment was achieved.

The effect of mixing speed， time， and low speed mixing process after placing defoamer on strength and slump of self-compacting cement mortar were investigated. Test results show that slump of cement mortar mixed by continuously variable transmission （CVT） mixer is larger than that mixed by conventional cement mortar mixer under the same mix proportions. Compared with mixing process with low speed and short time， slump and strength of cement mortar corresponding to mixing process with high speed and longer time increase 10% and 7%， respectively. Process of low speeding mixing after placing defoamer has an appreciable increase in compressive strength， which make the 7 d strength of cement mortar is larger than 28 d strength of cement mortar without this process， when their previous mixing process are the same.

For joint localization of time difference of arrival （TDOA） and angle of arrival （AOA）， an improved African vulture localization algorithm based on quasi reflective learning mechanism and parallel mechanism was proposed. The improved African vulture introduced a quasi reflection mechanism to enrich population diversity and speed up convergence in the process of searching for the maximum likelihood fitness function of the location model and in the iterative process， which also balanced the exploration and exploitation capabilities to a certain extent； The parallel mechanism was introduced to guide another population through the optimal individual of one population， which speeds up the convergence speed and enhances the optimization performance. From the results， the improved African Vulture algorithm is compared with AVOA， IHHO， CSSOA， PIO and CASSA， and shows faster convergence speed， more accurate positioning accuracy and better stability in solving the benchmark function and positioning model.

To address the problem of high computational complexity and inability to remove incorrect atoms in the orthogonal matching pursuit （OMP） algorithm when high estimation accuracy is required， an improved OMP algorithm based on the improved grey wolf optimization algorithm is proposed. Firstly， a nonlinear convergence factor based on the sigmoid function is proposed to improve the grey wolf algorithm， and a dynamic weighting method is introduced into the position update strategy of the grey wolf algorithm. Then， the improved grey wolf algorithm is applied to the field of compressed sensing DOA estimation， and the atom matching process of the OMP algorithm is optimized by using the improved grey wolf algorithm， which reduces the computational complexity and running time of the OMP algorithm， and introduces a backtracking thought to improve the correctness of the algorithm. Finally， simulation experiments demonstrate that the proposed algorithm has higher estimation accuracy， faster operation speed， and stronger anti-noise ability compared to the original algorithm.

In view of the complexity and nonstandard representation of modern Tibetan text， which affects the performance of speech synthesis system， this paper proposes a Tibetan text standardization method with the characteristics of easy maintenance and scalability. Firstly，a deep analysis was conducted on the different manifestations of Tibetan marker symbols and non Tibetan special symbols from other languages in Tibetan texts， and the special symbols were classified based on different features. Secondly， according to the different types of induction， the writing rules for converting 15 special symbols into Tibetan language were respectively established. Finally， using 13 490 sentences as the experimental data， the effectiveness of special symbols and Tibetan syllables in the text is identified and tested through the Tibetan grapheme-to-phoneme conversion test， and the sentences containing special symbols are standardized by the method of rule matching. The experimental results show that the omission rate of Tibetan phoneme transcription before standardization was as high as 4.69%， but after standardization， the omission rate of phoneme transcription was reduced to 0.01%， and the standardization accuracy rate of Tibetan text reached 99%.

To enhance the perception of small objects， based on the F-PointNet network，the FDG-PointNet 3D object detection model is proposed by combining dense connection and Gaussian distance features. Gaussian distance features is fused as additional attention features， and it effectively solves the low accuracy of instance segmentation in the F-PointNet network and enhances the noise filtering in the point cloud view cone. Based on the characteristics that dense connection can enhance feature extraction， the dense connection is used to improve PointNet++ network and enhance point cloud feature reuse. It alleviates low degree of feature extraction and gradient disappearance for small objects in the feature extraction process， and improves the accuracy of 3D object bounding box regression. The experimental results show that the proposed algorithm outperforms the benchmark method F-PointNet in three levels （easy， moderate， and hard） for the detection of car， pedestrian， and cyclist， which can achieve the average detection accuracy of 71.12%， 61.23%， and 55.71% for car， pedestrian， and cyclist at moderate level. It has the most significant improvement for pedestrian detection， and can increase 5.5% and 3.1% at easy and moderate levels， respectively. In summary，compared to F-PointNet algorithm， the proposed FDG-PointNet algorithm effectively solves the low accuracy of small objects detection and has strong applicability.

Aiming at the problems of large memory consumption， low equipment utilization， long training time and difficult convergence that occurred in the training process of the large-scale deep neural network （DNN） model， an adaptive parallel task scheduling method for the large-scale DNN model was proposed. Firstly， a multi-iteration asynchronous parallel management mechanism for model parallel was established， the specific scheduling process of micro-batch units was controlled to realize rational model partitioning and allocation of computing resources， and solve the problem of gradient delay updating during asynchronous iteration. Secondly， a computing resource allocation mechanism was designed based on topology awareness to achieve the best matching between model training tasks and computing resources. Finally， the runtime scheduling strategy for computing resources and model tasks is designed to maximize the overlap between computation and communication in the training process of fine-grained deep learning model， and improve the utilization of computing resources. Experimental results show that， compared with the existing model parallel methods， the proposed scheduling strategy can make full use of the computing resources of each GPU， and improve the training speed of large-scale DNN models by 2.8 times on average while ensuring the training accuracy of the model.

In order to effectively recommend online book resources to users， a recommendation method for online book resources based on deep fusion of interest information is proposed. Firstly， network users are clustered using the K-means clustering algorithm， and the accuracy of user clustering is improved by optimizing the clustering centers. Secondly， the degree of users' interest in book resources is calculated， and candidate recommended resources are determined based on this. Finally， an LSKGCN model is established to deeply fuse the long-term and short-term interests of users， score the book resources， and the online book resources with high scores are recommended to users. The experimental results show that the recommendation results of this method have high category diversity and content diversity.

In order to reduce the losses caused by APT attacks on the network and improve the security of network operation， it is necessary to control access in the network. A dynamic access control algorithm for multi domain interoperability under APT attacks is proposed. Firstly， an APT attack model is constructed based on Petri nets； Secondly， by calculating the trust experience， knowledge， and recommendations of each domain in the network under APT attacks， the trust values of each domain are obtained； Finally， the domain trust level is calculated based on the domain trust value， and the access permissions assigned to the corresponding roles in the domain according to the level are used to complete dynamic access control for multi domain interoperability. The experimental results show that the algorithm has high control efficiency and good control performance.

Aiming at the problems of missing details， dim color and reduced brightness in the images collected by the image acquisition system in the foggy scene， a multi-scale encoding decoding neural network （MSAOD） model was proposed based on the AOD theory for image dehazing. The proposed network model was divided into three modules. The first module is the preprocessing module， which divides the input image into two parts for preprocessing. The second module is the backbone module， which extracts the features of the output of the first part through the multi-scale encoder decoder. The third module is the post-processing module， which maps the feature map. The experimental results show that this method is superior to the mainstream deep learning and traditional methods in image dehazing effect， and the image after dehazing is optimized in detail， color， brightness and so on.

The collection of information features is an indispensable step in the process of studying human motion. In order to further optimize the problems existing in the current collection methods， such as poor collection effect， low collection accuracy， and long collection execution time， this paper proposes a method of human motion information feature collection based on digital portrait. Firstly， this method uses digital portrait technology to obtain landmarks of human motion video image， and uses improved background difference algorithm to extract human motion contour with landmarks as the center. Secondly， the epipolar constraint method and gray cross correlation algorithm are used to process the motion contour， and the human motion trajectory is obtained. Finally， the dynamic time warping algorithm is used to obtain the motion type， speed and other related information of the trajectory， and complete the collection of human motion information features. A group of human motion image data was randomly selected to verify the proposed method. The experimental results show that the proposed method has good information feature acquisition effect， high acquisition accuracy， and low acquisition execution time. The method of human motion information feature collection based on digital portrait can effectively obtain human motion information and realize human motion information feature collection， which is of great significance for improving the quality of human motion information feature collection.

Inertial navigation system will encounter various disturbances in the actual environment， such as vibration， temperature change， external magnetic field， etc.， and it is easy to accumulate large errors under long-term operation， making the stability and accuracy of the navigation system low. For this reason， a third-order non-static damping algorithm for long endurance inertial navigation under the constraint of periodic oscillation suppression is studied. Kalman filter is used to estimate the periodic oscillation error， error compensation mechanism is introduced to suppress the periodic oscillation error， fuzzy control is used to realize damping control， navigation parameters are modified， and the research of third-order non-static damping algorithm for inertial navigation is completed. The experimental verification shows that： The error of navigation parameters under the application of this algorithm is relatively smaller. With the passage of time， the error suppression effect of the algorithm is gradually enhanced， and the error is smaller and smaller. The position error is reduced to 0.15 m， the speed error is reduced to 0.10 m/s， and the attitude error is reduced to 0.06°. It shows that the algorithm provides an effective solution to improve the accuracy and stability of long endurance inertial navigation system.

Aiming at the problem that the battery defect detection is susceptible to black appearance interference， which leads to the limited detect identification of the observation image under a single light source， an end-to-end deep-learning photometric stereo network （DPSNet） is proposed. Firstly， the photometric stereo feature generator （PSFG） enables model to facilitate transformation from multi-input features to normal feature so that better excel in defect signals. Then channel co-attention （CCA） module explores the channel interactions between each input and surface normal for informative representations， and fuses modify features to coordinately enhance global representation. Finally， spatial pyramid pooling （SPP） and feature pyramid networks （FPN） achieve multi-scale prediction. The experimental results on the self-built Battery101 dataset show that the proposed method achieves better effect. In addition， the ablation experiment further verifies the effectiveness of each module in the model.

To address the issues of particle degradation and the difficulty in selecting the importance density function in particle filtering， a Constrained Optimization-Based Adaptive Fading Memory Square Root Mixed-Degree Spherical Simplex-Radial Cubature Particle Filter （COAFM-MSSRCPF） algorithm was proposed. The advantages of constrained optimization， adaptive fading memory， square root filtering， and Mixed-Degree Spherical Simplex-Radial Cubature Kalman Filtering （MSSRCKF） are combined in this algorithm. By employing the Mixed-Degree Spherical Simplex-Radial sampling criterion， higher accuracy compared to traditional Cubature Kalman Filtering （CKF） and lower computational complexity than High-Degree Cubature Kalman Filtering （HCKF） are achieved. The adaptive fading memory square root strategy is utilized for predicting and updating the covariance matrix， with the weight of current measurement information being enhanced and the influence of historical data reduced. As a result， issues of covariance matrix asymmetry， negative definiteness， and filter divergence are avoided. The noise covariance matrix is dynamically adjusted， and the convergence speed and accuracy of state estimation are improved by constraining the ratio of error covariance to measurement noise covariance. Simulation results demonstrate that the COAFM-MSSRCPF algorithm effectively suppresses filter divergence in SINS/GNSS integrated navigation systems. Filtering accuracy and robustness are significantly improved compared to the Fading Memory Cubature Particle Filtering （FMCPF） and traditional Cubature Particle Filtering （CPF） algorithms.

In order to address the identification of the quantized Wiener system， a repetitive gradient learning identification algorithm was proposed. Firstly， based on the decomposition technique， the quantified Wiener system was transformed into an identification model with parameter separation， in which the computational burden of was reduced. Secondly， the observation data were extended using data window theory to obtain more quantitative system modal information. To address the issue of moving time window length， the idea of repetitive learning was integrated into the parameter adaptive law update mechanism， which greatly improves the estimation performance. Finally， the convergence of the estimator and the comparisons of example have used to show the effectiveness and advantages of the proposed algorithm.

In order to improve the power supply reliability of the distribution network system， solve the current problems of insufficient power supply and high production cost， an annual power balance analysis was conducted for the distribution network system with hybrid power sources. First， the scene tree generation method was used to sample and select the wind power， photovoltaic， and load data within the year. Three typical days were selected in each month according to the probability principle to reflect the characteristics of wind and wind power generation and load throughout the year. Secondly， considering the operating constraints of wind and solar power plants， gas turbine units， batteries， and continuous and discrete reactive power compensation devices， an optimization model of the distribution network system was established to minimize with the goal of minimizing the annual operating cost. The optimization solution problem is a second-order cone programming model. The Yalmip optimization solution tool was used to model the distribution network system， and the Gurobi solver was used to solve the power balance problem of each typical day. Based on the solar radiation， temperature， wind speed， power load of Chengdu， the design example of the IEEE 33-node distribution network system is used to finally realize the production capacity distribution and economic analysis of the annual power balance.

At present， the detection algorithm cannot accurately detect the deformation of the main material of the power tower body. Therefore， a carrier phase difference based deformation detection algorithm for the main material of the power tower body was proposed. Firstly， combining local mean decomposition method and support vector regression machine to detect and repair cycle jumps in satellite signals.Secondly， a carrier phase differential detection model was established， and during the deformation detection process， the detection model was updated through Kalman filtering. Finally， the LAMBDA algorithm was used to calculate the integer ambiguity in the carrier phase differential detection model， and the calculation results are substituted into the model. The updated carrier phase differential detection model is used to detect the deformation of the main material of the power tower body. The experimental results show that the proposed algorithm has high cycle slip detection accuracy， good repair effect， and high deformation detection accuracy.

A trajectory tracking control method based on adaptive model predictive control （MPC） was proposed to address the influence of model parameter uncertainty and gust disturbance on the flight trajectory of biplane flapping-wing micro air vehicles （BFMAV） during its task process. Firstly， a six degree of freedom nonlinear dynamic model was built according to the structural characteristics of a BFMAV. Secondly， the internal and external disturbance were introduced to characteristic the model parameter uncertainty and gust disturbance， the state equation of the BFMAV was further determined， and the attitude guidance law was designed to adjust the attitude angle in real-time. Thirdly， a BFMAV trajectory prediction model considering model-environment uncertainty was built by embedding the internal and external disturbance into the MPC model. Finally， different control methods were used to simulate the trajectory tracking states in the free flight and with internal and external disturbances to verify the effectiveness of the proposed method. The results show that the proposed method has stable tracking performance and small tracking error compared with the traditional MPC method. In the case of model uncertainty and external disturbance， the trajectory tracking task of the BFMAV can be completed well.

In order to maintain a stable dynamic posture during the operation of biped walking robots and complete work tasks with high efficiency， a motion differential equation based dynamic posture stabilization control method for biped walking robots was proposed. Firstly，establish a biped walking robot model， determine its coordinates in the world coordinate system and rigid coordinate system， and express them in the form of homogeneous coordinates； Secondly，the dynamic attitude stability controller of biped walking robot is constructed， and the differential equation of motion is introduced to constrain the dynamic attitude of the robot， so that it can remain stable in any constrained surface；Finally，compared with the other two algorithms， the experimental results show that the proposed method is highly feasible and reasonable. Compared with other algorithms， the proposed method has better self-balancing ability， anti-interference ability and dynamic attitude stability control ability.

In order to improve tracking accuracy of intelligent vehicle under non-linear reference trajectories， a curvature augmentation model predictive control/propotional integral controller with front wheel angle compensation was proposed. First， based on tracking deviation vehicle model， model predictive control algorithm was designed by using curvature as augmentation state， and influence of curvature broadening was analyzed. Then， Lyapunov direct method was used to obtain predictive horizon that ensures algorithm stability， and system steady-state error was calculated. To eliminate steady-state lateral deviation， propotional integral controller was designed to compensate front wheel angle. Finally， simulation was conducted and results show that， the designed controller improves trajectory tracking accuracy and achieves better convergence rate， while ensuring stability and smoothness， and obtains good control effect under limit conditions.

A dynamic control strategy for variable speed lane change trajectory tracking was proposed. It includes a double PID longitudinal speed controller and a lateral model predictive controller using comprehensive evaluation indexes： lateral error， yaw rate， and side-slip angle. coefficients of each evaluation index based on the estimated tire-road friction coefficient and predicted vehicle status information. This allows achieving dynamic control between tracking accuracy and lateral stability. To verify the effectiveness of the algorithm， variable speed lane changing scenarios with high and low tire-road friction coefficients were designed， and CarSim and Simulink simulations were conducted. The simulation results demonstrate that the proposed dynamic control strategy realizes the dynamic adjustment of weight factors under different tire-road friction coefficients. It ensures smooth control output during lane changes， improves trajectory tracking control accuracy and lateral stability of unmanned vehicles simultaneously， and exhibits good coordination control effect.

In response to the inefficiency in the mechanized segmented harvesting of rapeseed and the low utilization rate of dedicated chassis， and combining with the advantages of the large inventory of track-type power chassis of combine harvesters in China， a wide-foldable rapeseed windrower based on a track-type chassis was designed. Adopting a modular design approach， the structure and the working operations of the windrower were explained. The combine harvester could be adjusted to be a windrower by just switching the header， and the windrower could meet the requirements of road transportation once the header was folded. The header was divided into two bilateral symmetrical parts. During the folding， both of the parts will rotate firstly and then contract to the center of the header. Based on the dynamics and kinematics， the relationship between the folding position parameters of the header and the load of the folding hydraulic cylinder was analyzed. The rotation point of the header was determined to be 950 mm from the inner side of the frame， and the contraction stroke was 700 mm. Furthermore， the structure and working parameters of the wheel， cutter and conveying device were analyzed and determined. The load and constraint of the header frame and the hydraulic folding frame were clarified. The static analysis of the unfolding and folding status of the header and the topological optimization structure design were carried out. With the minimum stress and strain as the optimization goal， the engineering scheme of the initial model material removal rate of the header frame （85%） and the hydraulic folding frame （80%） was determined by comparative evaluation. At last， the field experiment was conducted to evaluate the functions of the windrower. The results of field experiments showed that the header could fold and work smoothly. The average laying angle of the rape was 24.4°， and the difference of laying angle was 8.52°. All the indicators could meet the field transfer requirements of the windrower. This study could provide reference for the structural design and optimization of folding header for harvesting equipment.

To address the problem of existing regression analysis models being prone to estimation distortion or difficult to estimate accurately， an HM-OLS stepwise regression analysis model was proposed. First， the data was compressed and reduced in dimension using principal component analysis （PCA） to solve the problem of multicollinearity. Second， the processed data was used to construct a matrix， and the model parameters are estimated using least squares method， completing the model fitting， and conducting heteroscedasticity test and multicollinearity detection. Finally， the AIC value was used as a reference， and the forward stepwise regression method is used to select influencing factors， re-fitting the model， and completing the correlation analysis. The experimental results show that the HM-OLS stepwise regression analysis model proposed in this paper effectively eliminates the problems of scale differences and multicollinearity， and its stability and fitting effect are significantly better than those of traditional OLS and ridge regression analysis models. It can also accurately analyze the influencing factors with strong correlation to student academic performance in the network learning space.