Aiming at the problems of over-redundancy of low-quality data， difficult mining of feature data and time-consuming manual recording in the traditional data acquisition and analysis process， a driving behavior data filtering method based on dynamic time regularization algorithm under man-machine co-driving was proposed. Firstly， a dynamic time warping algorithm model was built in Python environment to realize real-time deviation calculation of two sequences by means of rolling time window. Then， considering the statistical characteristics of different distance calculation methods， the deviation threshold of triggering records was designed， and the model was optimized with the global constraint of structured path. Finally， simulation analysis and real vehicle test were conducted to compare the data filtering methods under different constraints. It was found that the proposed filtering method based on Sakoe-Chiba constraints can automatically filter out an average of 53.15% of invalid data in the data preparation stage， saving 1.87 TB of data storage space per hour， the effectiveness and feasibility of the proposed method is verified.

An extended-range electric vehicle model with a compressed natural gas （CNG） engine was established referring to the experimental test data of the CNG engine. A rule‐based method for power split was proposed to ensure that the CNG engine works in the high-efficiency area. However， its current fluctuates greatly， which will result in a damaged battery. In order to suppress the current surge and further improve fuel consumption， a model predictive control （MPC） method for power split coordinating fuel consumption and battery protection was proposed. Simulation results show that the proposed MPC method can further reduce fuel consumption by 3.7% and suppress the current surge for effectively protecting the battery， as compared to the rule‐based method under a world light test cycle （WLTC） driving cycle.

Aiming at the problems of actuator nonlinearity， parameter uncertainty and strong model dependence in the existing hydraulic suspension systems of rescue vehicles， a control method of the hydraulic active suspension system-model predictive control method with an extended state observer was proposed. Firstly， the vehicle's pose information was obtained through the on-board inertial navigation system and used to solve the expected displacement output value of each suspension actuator based on pose deviation method. Secondly， dynamic model of the hydraulic suspension system was completed for rescue vehicles， and the nonlinear disturbance and unknown output signal in the system were estimated through an extended state observer. In view of the limitation of the output for the suspension system， a model predictive controller with an extended state observer was proposed to effectively track the expected displacement value of each hydraulic actuator. The vehicle test platform with hydraulic suspension system was built to verify the effectiveness of the control method， and some road tests were carried out by comparing the passive suspension and traditional PID control methods. The research results show that， compared with the passive suspension and the traditional PID method， the proposed active suspension control method reduced the root mean square value of the vertical displacement by 35%， the pitch angle by 17%， and the roll angle by 23%， which significantly improved the ride comfort and handling stability of the vehicle.

In order to solve the intelligent vehicle trajectory tracking controller based on model predictive control with long processing time and low real-time performance， a model predictive trajectory tracking control method based on variable step size in the control time domain was proposed using matrix-blocking strategy. The matrix-blocking method was used to change the control time domain step size and integrated into the quadratic programming solution process of model predictive control， and the objective function and system constraints were reconstructed to reduce the number of variables to be solved in the optimal control sequence during the solution process， and the calculation time are also be reduced. Based on Simulink and CarSim co-simulation platform， the proposed method was compared with the traditional model predictive control method. The results demonstrate that compared with the traditional model predictive control method， the proposed method not only reduces the average calculation time by 24.39% and the maximum single calculation time by 45.05%， but also ensuring the trajectory tracking accuracy. The performance of the controller using the dense before sparse block matrix is better than the average block matrix.

Based on the self-developed hydraulic variable valve train， the improvement potential of the internal exhaust gas recirculation （EGR） coupled ignition strategy for methanol engine energy saving and emission reduction was studied. The results show that the exhaust valve lift decreases and the internal EGR increases. An increase in internal EGR results in a decrease in the peak cylinder pressure and in-cylinder temperature， a delay in the combustion center， an increase in the combustion duration， and an increase in the variation of the combustion cycle. Properly adjusting the ignition timing， the methanol engine can still distribute the combustion center within a reasonable range and run smoothly even at 42% internal EGR. In addition， internal EGR combined with ignition timing has different effects on improving fuel consumption and emissions under different loads. The effective thermal efficiency and fuel economy are the best at 20% load， which are improved by 3.2% and 12.7% respectively. The NO _x and HC_{emission was the best at 40% load， which decreased by 64.25% and 20.8%， respectively.}

The electric heating spray cooling drag reduction test bench was used to systematically study the influence of spray flow rate， spray inclination angle， airflow temperature， and airflow speed on the drag reduction performance of the engine exhaust system. The reason why spray cooling can significantly reduce drag was revealed by DPM-based numerical simulation. The research results show that the drag reduction rate increases with the spray flow rate when the spray flow rate is between 12 mL/s and 24 mL/s. When the spray inclination angle is increased from 45° to 90°， the drag reduction first remains unchanged and then decreases. The drag reduction effect of 45° and 60° spray inclination is the best. When the airflow temperature is between 200 ℃ and 300 ℃， the drag reduction rate increases with the airflow temperature. The drag reduction rate decreases gradually when the airflow velocity increases from 20 m/s to 30 m/s. The reason for the significant reduction in the drag of the exhaust system after spraying is that the airflow velocity reduces， the turbulent intensity reduces， and the local and frictional drag reduces.

The fatigue crack growth behavior of SMA490BW steel welded joints for high-speed train bogies under different stress ratios in different welding areas was studied， and the microstructure and fracture morphology of the welded areas were characterized by optical microscope and scanning electron microscope. The results show that under the same stress ratio， the different welding areas （weld nugget zone， heat affected zone） of SMA490BW steel have little effect on the crack growth rate. The stress ratio has a greater influence on the crack growth rate. Under the same stress intensity factor， the crack growth rate at R=-1 is significantly higher than that at R=0； the stress ratio has a greater influence on the c parameter in the Paris formula and less on the m parameter.

The analytic hierarchy process （AHP） was presented to realize decision-making of welding positions during the horizontal multipass welding process with T-joints. First， the feature points of the groove were detected with visual sensing technology， and the distances between other feature points and the rightmost one respectively in the X/Y direction were used as factors to build the AHP model. Second， a computational method that integrated welding experience related to angular distortion control was proposed to automatically determine the elements in the comparison matrixes. Finally， the decision-making result of welding positions was conducted using the maximum posteriori weight principle. The experimental results on multiple decision-making processes with the webs of different thicknesses showed that the proposed method owns strong anti-interference ability and the accuracy is over 90%.

In order to improve the thermal conductivity and electrical insulation properties of polyimide matrix composites， fluorographene nanosheets were prepared by ultrasonic exfoliation process， and fluorographene/polyimide composite films with horizontal directional thermal conductivity characteristics were prepared by hot-pressing induced orientation process. The results show that increasing the amount of fluorinated graphene can improve the in-plane thermal conductivity of the composite film. When the mass fraction of fluorinated graphene is 30%， the in-plane thermal conductivity of the composite film is 2.4 W/（m·K）. When the amount of fluorinated graphene increases， the DC withstand voltage performance of the composite film slightly decreases， but the electrical tracking resistance of the composite film increases due to the improved phonon transfer efficiency in the composite film. The above research results can provide reference for the preparation of composite films with high thermal conductivity and high electrical insulation performance.

For seeking cheap， high performance and stable catalysts to replace noble metals. In this paper， sulfide/phosphide catalysts were obtained by one-step annealing treatment with transition metals Ni/Co， and the successful preparation of the catalysts was verified by electron detection and elemental analysis method. The electrochemical test results showed that the catalytic activity and charge transfer rate of the prepared Ni-Co sulfide/phosphide exceeded those of Ni-Co hydroxide/sulfide/phosphide due to the synergistic effect between anions/cations. Ni-Co sulfur/phosphide only requires an overpotential of 106 mV to achieve a current density of 10 mA·cm^-2 and the stability was more than 20 h.

In order to improve the safety of drivers taking over intelligent vehicles under different risky traffic scenarios， the characteristics and differences of drivers' taking over behaviors under three typical risky environments， namely highway， mountain road and city road， were studied in a simulated driving platform. The test collected and analyzed the driving behavior data of 16 drivers in three types of typical risky environments. The results showed that： ① The emergency actions taken by the drivers when taking over the vehicle included braking， steering and simultaneous steering and braking， with the percentages of each action being 71.4%， 16.7% and 11.9% respectively. ② There was no significant difference （P>0.05） in the reaction time of driver taking over in different dangerous traffic scenarios； There was a significant difference in the maximum longitudinal deceleration between highway and city road （P=0.002<0.05）， and a significant difference in the maximum longitudinal deceleration between mountain road and city road （P=0.048<0.05）. There was a significant difference in the maximum steering wheel Angle between city road and mountain road （P=0.015<0.05）， and there was a significant difference in the maximum steering wheel Angle between mountain road and highway （P=0.000<0.05）. It shows that different takeover measures should be used to improve the safety and effectiveness of takeover in different risky traffic scenarios. The results of the proposed method can provide reference for the development of intelligent vehicle takeover systems in different risky environments.

Combing the existing public transportation lane design specifications around the world， it is concluded that the setting of bus lanes usually considers the peak hour road bus passenger volume of section， bus traffic volume of sections， road conditions and other objective indicators， with little consideration given to the operational stability of bus routes on the road and its influence of improving passengers' experience of bus service reliability. In the post-operation evaluation of the public transportation lanes， operational reliability and efficiency assessment are finished regularly. Due to the lack of conditions for setting up dedicated bus lanes， some dedicated bus lanes have poor operational effects and no improvement in passenger experience. Therefore， considering passenger experience， a potential public transportation lane demand identification method based on the stability of bus operation was proposed. Based on the actual road conditions and bus vehicle GPS data， the deviation of bus vehicles from the planned schedule on each road section or station was analyzed， and two indicators， namely the bus congestion index and bus delivery time volatility were constructed， to form a classification method based on bus operation stability of road sections or stations. The proposed method was applied to the roads of 69 bus routes in central Shanghai， and the identified roads with poor bus operation stability were considered as potential needs for public transportation lanes. The method can also target public transportation lanes that need further optimization and improvement.

Focusing on the lane-changing decision-making behavior of autonomous vehicles， the microscopic lane-changing behavior of vehicles was studied based on the molecular dynamics theory. On the basis of objectively quantifying the intention to change lanes, a molecular dynamics model of lane changing decision-making behavior was further established by introducing the interaction potential between vehicles. Then， the relationship between the initial moment of lane-changing and the completion moment of lane-changing， as well as the dynamic influencing factors of lane-changing were systematically analyzed. The impact of microscopic lane-changing behavior on macroscopic traffic flow was explored. Finally， with SUMO software， the SL2015 lane-changing model and the molecular dynamics lane-changing model were used for simulation experiments and comparative analysis. The results show that the molecular dynamics lane-changing model has good safety， stability and practicability. The molecular dynamics modeling of lane-changing decision-making behavior for autonomous vehicles comprehensively considers the dynamic influencing factors in the traffic scene， which can more objectively and reasonably demonstrate the lane changing behavior characteristics of autonomous vehicles.

To meet the demand of direct passenger travel within the scope of urban agglomeration， the through lines plans based on the connection of multi-level rail transit were proposed， and the gain effects of passenger transport system brought by different plans were discussed. The specific contents are as follows： a super network model of urban agglomerations multi-mode transportation was constructed based on the theory of hyper network. The generalized travel cost of passenger super path was quantified. A bi-level programming model for the optimal setting of through lines was constructed， in which the upper level programming considered the profit of the transporter and the saving of the passenger travel cost brought by the setting of through lines. The lower level programming was a balanced allocation model for the multi-mode transportation system of urban agglomeration. Genetic algorithm and successive average method were used to solve the upper and lower level models. Finally， the passenger lines connecting the main hubs in Beijing-Tianjin-Hebei Urban Agglomeration were taken as an example to illustrate the performance and applications of the model， and the rationality of the scale of the through line was demonstrated. The results show that， there are 11 alternatives through lines between Beijing Capital International Airport， Beijing railway station， Beijing West Railway Station， Beijing South Railway Station and Tianjin railway station. The best through lines are No. 1， 2， 5， 6， 7and 9. The reasonable scale of the direct line is 3 at present. After opening， the system benefit will increase by 28.03% and the average travel time of passengers will reduce by 10.02%.

In order to effectively predict crash on elevated expressway， taking Yan'an elevated expressway in Shanghai as the research object， based on its traffic flow and crash data， an improved Gene Expression Programming algorithm with additional elite gene bank and extinction mechanism was applied to dig out ‘Crash Prediction Empirical Formula’. The prediction accuracy and interpretability of the empirical formula were verified by comparing with the results of machine learning and statistical analysis. The crash of another expressway was predicted by empirical formula without retraining and calibration， and the portability of the empirical formula was verified. The research results indicated that the prediction performance of the empirical formula on the Yan'an elevated expressway dataset is significantly improved compared with the traditional Logistics regression， and the receiver operating characteristic curve area and F1-score indexes are consistent with the artificial neural network model， identifying 74% of the crashes correctly. The good performance of the empirical formula on Hangzhou Shangtang elevated expressway dataset shows that the empirical formula has basic portability. In conclusion， the gene expression programming algorithm considers both high accuracy and interpretability for the crash risk prediction problem， and shows portability， which is helpful to establish a low-cost and efficient crash prediction system.

In order to analyze and quantitatively evaluate the effectiveness and safety of drivers' intervention behavior in autonomous truck road testing， the characteristic parameters were constructed with the truck speed， acceleration and angular velocity as the operational indicators. Autoencoder （AE） and K-means++ algorithm were combined to build a composite mode to identify five intervention operational patterns， including sharp deceleration， sharp acceleration， transverse right turn， transverse left turn and transverse and smooth intervention. Combining multiple factors of human-vehicle-road-environment when drivers took over and intervened， the intervention reasons were divided into three categories： risk-avoiding intervention， trust intervention and optimized intervention. According to different operational patterns and characteristics variation with time， intervention and control time was analyzed， and the reasons for takeover were identified. Based on autonomous truck road testing data， the 778 intervention samples were extracted to evaluate and verify the rationality and effectiveness of the method. The results show that， affected by factors such as the degree of scene risk， intervention and control time of sharp deceleration and smooth intervention operation pattern is significantly less than other intervention operating patterns in the autonomous truck road testing process； risk-avoiding intervention is mainly the sharp acceleration/deceleration intervention operation pattern， trust intervention is mostly manifested as transverse right/left turning and smooth intervention operation pattern， and optimized intervention only has sharp acceleration intervention operation mode； the types of risk-avoiding intervention and trust intervention account for more than 80%， indicating that drivers still take ensuring the safety of road testing as the primary goal in autonomous truck road testing.

Aiming at the problem that the existing network is difficult to accurately recognize the target of road detection and vehicle distance in real time， a vehicle distance warning method based on improved YOLOv4-tiny algorithm was proposed. Firstly， the feature extraction structure of YOLOv4-tiny algorithm was summarized， and the shortcomings of the original network structure were analyzed. Secondly， SPPF of spatial pyramid pooling layer was added to the original network to further extract target features and enhance the ability to express deep semantic information. Feature Pyramid Network （FPN） structure was added with down-sampling channel and CSPnet layer to fully integrate multi-scale image features and avoid the loss of shallow information. Finally， the Mosaic data enhancement method was used to enrich the data set training samples， and the improved YOLOv4-tiny algorithm was combined with the principle of single visual distance detection. The vehicle distance warning experiment was conducted by setting three levels of information cues according to the vehicle distance. The results show that the detection speed of the proposed algorithm on PASCAL VOC dataset is 43 frames/s， and the average accuracy is 81.25%， which is 3.59% higher than that of YOLOv4-tiny algorithm. It can be seen that the improved YOLOv4-tiny algorithm has good target detection accuracy while meeting the real-time requirements of detection， which has guiding significance for improving the application effect of vehicle distance warning method.

To address the issue of insufficient feature extraction and imprecise target localization in existing algorithms for detecting bridge bolt defects due to the complexity of bolt backgrounds and their small size， a bridge bolt defect recognition method based on enhanced YOLOv5s was proposed. Attention mechanisms in the backbone network was introduced to enhance the model's ability to extract bolt features and deepen its focus on global bolt characteristics. The spatial pyramid pooling structure was optimized to reduce the loss of bolt feature information. MPDIoU was employed as the bounding box regression loss function to improve the accuracy of bolt bounding box regression. The YOLO detection head was decoupled to eliminate the adverse effects of shared detection head in target detection on the regression of bounding box positions. Training and testing were conducted on 3810 self-made bolt image datasets of four typical defects： bolt rusting， bolt loosening， bolt detachment， and nut detachment， as well as normal bolts. Experimental results show that the algorithm achieves a detection accuracy of 90.8% for bolt defects， which is a 3% improvement over YOLOv5s， and a mean average precision of 92.6%， representing a 4.3% improvement over YOLOv5s. This method can be applied for intelligent recognition of bolt defects in bridges.

In response to the challenges of significant randomness and susceptibility to local optima in the sparrow search algorithm， an enhanced approach was proposed integrating multiple strategies. During the initialization phase， a good point set strategy was introduced to ensure population diversity and thorough exploration. The discoverer's position update incorporates a dynamic learning mechanism， effectively balancing global optimization and local exploration capabilities. Simultaneously， the follower's position update integrates a Lévy flight disturbance mechanism， reinforcing local escape capabilities. Finally, the proposed method was applied to solve the coverage problem of wireless sensor networks. Through a multi-objective coverage optimization function， considering coverage rate maximizing， redundancy minimization， and energy consumption equilibrium maximizing. The simulation results show that the three improvement measures significantly improve the algorithm performance, enhance the coverage quality of network nodes, and effectively improve the overall performance of the network, proving that the proposed method has good performance in practical applications.

To enhance the node localization accuracy and stability of the tradition three-dimensional DV-Hop （3DDV-Hop） algorithm in wireless sensor network （WSN）， a 3DDV-Hop positioning optimization algorithm based on the improved multi strategy sparrow search algorithm （MISSA-3DDV-Hop） was proposed. Firstly， a communication radius classification method was used in the anchor nodes to refine the hop value of positioning nodes， which improves the accuracy of the calculation of the minimum hop number. Then the weighted average of hop distance error and estimated distance error was applied to correct the average hop distance between nodes， so as to reduce the distance estimation error between anchor nodes and unknown nodes. Finally， the sparrow search algorithm was used to optimize the location of unknown nodes in 3DDV-Hop algorithm， and the good-point set and discoverer-follower adaptive adjustment strategy were introduced. By establishing the error fitness function and objective function， the distribution and diversity of the initial population， global convergence speed and local optimization ability of sparrow search algorithm were enhanced. The simulation results show that the improved algorithm effectively improves the positioning accuracy and convergence speed compared with the traditional 3DDV-Hop， IPSO-3DDV-Hop and IGA-3DDV-Hop algorithms.

A multi-input underwater image recovery method based on TransFormer and convolutional neural network （CNN） was proposed to address the issues of low contrast， poor detail representation and color error in underwater images.TransFormer and relative total variatio were used to construct a depth feature extraction module to fuse the texture map extracted by relative total variatio （RTV） with the image information extracted by TransFormer， which effectively enhances the detail features of the image. The color correction module was constructed by using automatic color equalization and Lab color space to enhance the image contrast and correct the color. A multinomial loss function was used to constrain the network convergence to obtain the enhanced clear underwater images. Finally， the quantitative and qualitative comparative analysis of the proposed method with other methods on the test set was carried out， and the experimental results show that the images processed by the proposed method outperform other comparative methods in terms of sharpness， color performance and texture information.

In fully cooperative tasks， the MADDPG algorithm has credit assignment and poor stability of training problem. To address this problem， a LSTM-MADDPG multi-agent cooperative decision algorithm based on asynchronous collaborative update was proposed. According to the idea of Difference Reward and Value Decomposition， LSTM was used to extract the characteristics between trajectory sequences. The global reward division was optimized to realize the agent's reward distribution. In order to meet requirements of algorithm joint training， the high-quality training set was constructed. Then， the asynchronous cooperative update method was designed to joint train the LSTM-MADDPG network， and realize the cooperation of multi-agent. In cooperative capture scene， the simulation results show that the convergence speed of the proposed algorithm is increased by 20.51% compared with the QMIX. After the convergence of algorithm training， the update method of asynchronous cooperation reduces the mean square error of normalized reward value by 57.59% compared with synchronous update， which improves the stability of algorithm convergence.

Using knowledge graph technology to improve the quality review method of building information model， the BIM model of power grid engineering was selected as the research object， and a BIM model inspection system based on knowledge graph was developed to verify whether it meets the relevant specifications and standards of power grid engineering. By using the BIM model of a certain electrical engineering project as a case study， an inspection report was generated， and the results verified the feasibility and practicality of this method. This technology provides a new direction for the deep integration of architectural design and information technology.

To prevent information theft in the big data environment， a mobile Agent based identity authentication method for wireless power mobile terminals was proposed.The mobile Agent was used to manage the distributed key of power mobile terminals， exchange heterogeneous network data and resources in the mobile Agent， and obtain the node matrix table to obtain the key required by users， the system member trust model based on Agent is constructed to ensure that the mobile terminal identity authentication is in a good operating environment. According to the user access results obtained by the mobile Agent， the user access dynamic key is generated under the irreversible calculation method， and the user authentication information is encrypted twice using the obtained authentication code to achieve the identity authentication of the wireless power mobile terminal. The experimental results show that the proposed method has the advantages of short key， high accuracy of identity authentication and small amount of authentication computation at the same encryption level.

In order to solve the problems of uneven distribution of companion stars during the construction of radial features and unstable feature values in the matching process with circular features in the traditional radial and circular feature recognition algorithms， a star identification algorithm combining equal-area circular division and companion star vector pinch angle was proposed. By choosing the dynamic radius， so that the area of circles under radial feature is equal， the probability of companion star falling on each circle was guaranteed to be equal， which effectively improves the recognition rate of radial matching process. For the redundant navigation stars after radial matching， the vector pinch features composed of companion stars were used for screening， while the one with the highest matching feature value was selected as the final identification result. The results of experimental simulation show that the success rate of star identification reaches 95% when the added position noise is 1 pixel； the recognition rate reaches 98% when the number of false stars is 2； compared with the traditional radial and circular recognition algorithms， the proposed method has obvious advantages in the recognition rate and recognition speed.

A novel data-driven learning method to achieve static output-feedback robust control of unmatched dynamic systems was proposed， which uses the techniques originally developed for optimal control. The robust control was first transformed into the optimal control of an augmented system， taking unmatched dynamics into consideration. Then， to design the output-feedback optimal control， an output-feedback algebraic Riccati equation was derived by tailoring its state-feedback control counterpart. Once more， an adaptive online learning method was designed to avoid using the observer， where two operations （i.e.， vectorization and Kronecker's product） were adopted to reconstruct the output-feedback algebraic Riccati equation. Finally， the required persistent excitation condition was further relaxed to realize the rapid convergence of the estimated parameters. Simulation results show the effectiveness of the proposed control method and learning algorithm.

In order to improve the transmission security of millimeter wave in radio frequency （RF） transmission， the research on 5G millimeter wave RF algorithm in multi-input multi-output （MIMO） radar search mode was proposed. Based on the transmission characteristics of 5G millimeter wave, the radar interception factor of MIMO radar search mode was obtained. A millimeter wave radio frequency stealth model was established and the objective function was obtained. The radio frequency interception probability of the radar search mode was calculated. The best estimation value of millimeter wave RF channel was obtained by combining the millimeter wave channel sparsity results with Bayesian algorithm， from which the best value of RF channel was calculated， and the RF transmission was realized by using the estimated signal transmission after stealth. The experimental results show that when the millimeter wave RF transmission time is 140 s， the ρ of the proposed method is as high as 0.83. It is proved that when this method is used for millimeter wave RF transmission， the transmission quality of millimeter wave is high， the phase amplitude deformation is not easy to occur， and the high-quality transmission of 5G millimeter wave RF can be realized.

The crashworthiness of a biomimetic sandwich panel （BSP） was explored based on the shape of the shrimp chela. The finite element model was set up according to the material data obtained from the tensile test. A sample of BSP was fabricated， and used to verify the finite element model， which was reliable. The crashworthiness indexes of BSP-L0， BSP-L180， BSP-T0 and BSP-T180 were compared， and the BSP-T0 scheme was selected for further study. The influence of arc height ratio γ and trapezoidal cell element narrow bottom width B on the crashworthiness of BSP-T0 was analyzed， and it was obtained that γ has a greater influence than B on the crashworthiness of BSP-T0. Genetic algorithms were used for parameter optimization， an optimal solution of the Pareto frontier is SEA=3.080 kJ/kg， PF=30.700 kN， CFE=86.884%， the corresponding parameters are γ=0.355，B=4.23 mm， the errors of prediction and simulation results are all within 2%， and the prediction effect was ideal. The research results can provide new ideas for the design of sandwich structures.

Aiming to improve the accuracy of wet clutch pressure control of tractor power shift transmission （PST）， a model-free adaptive control method for wet clutch pressure based on improved genetic algorithm is proposed. Firstly， the mathematical model of the wet clutch hydraulic system was analyzed and established， and the clutch pressure controller was constructed based on the partial form dynamic linearization based model free adaptive control （PFDL-MFAC） algorithm. Then， the variable-proportion elite retention strategy， K-means clustering algorithm and catastrophe strategy were introduced to improve the genetic algorithm， and a PFDL-MFAC controller parameter tuning method based on sensitivity analysis and improved genetic algorithm was proposed. Finally， the clutch pressure control test based on the tractor transmission control unit （TCU） hardware-in-the-loop test platform was carried out. The results show that the improved genetic algorithm has better convergence speed and optimization accuracy. Compared with PID， PFDL-MFAC has faster clutch pressure response and better robustness， which can meet the requirements of tractors. At the same time， the shifting quality of the transmission based on the algorithm in this paper is better. The research results provide a basis for improving the shifting quality of power-shift tractors.