The optimal control method has become the mainstream research and industry deployment method for lateral motion control in autonomous driving. The LQR method is widely used due to its advantages of low online computational load and good real-time performance，but it cannot consider time-varying references and steering delay. The presence of delay can cause the LQR method to lose stability at high speed，so it is essential to solve this problem while maintaining the characteristic of small computational load of LQR. In this paper，under the premise of ensuring real-time performance，the problem of LQR's inability to consider time-varying references and steering delay is solved. By incorporating road curvature as time-varying references，steering delay characteristics as pure delay，and first-order inertial section into the tracking error state equation，and by looking up the KKT inverse matrix part corresponding to the control time domain into the real-time solver，the aim is to reduce computational load and ensure controller real-time performance. The simulation results demonstrate that the constructed EqLPV-MPC controller can effectively handle road curvature changes. Compared to the LQR method，under the condition of dual lane change at a speed of 72 km/h，the lateral error decreases by 39%，with the heading error decreasing by 52%，and the lateral deviation of the center of mass decreasing by 28%. The results from real vehicle tests show that under dual lane change conditions，the controller constructed in this paper can keep the maximum lateral error within 0.1 m.

With the development of Intelligent Connected Vehicle （ICV） technology，ICVs with limited computing resources face the problem of significantly increased computational demand. ICVs can offload tasks to Mobile Edge Computing （MEC） servers via Roadside Units （RSU）. However，the dynamic and complex nature of vehicular networks makes task offloading and resource allocation highly challenging. In this paper，it is proposed to minimize task computing energy consumption by controlling task offloading decision，communication power，and computing resource allocation under environmental and resource constraints. To address the coexistence of discrete and continuous control variables in the problem，a Hybrid Deep Reinforcement Learning （HDRL） algorithm is designed. The algorithm employs the Double Deep Q-Network （DDQN） to generate task offloading decisions and the Deep Deterministic Policy Gradient （DDPG） to determine communication power and MEC resource allocation. Furthermore，an Improved Prioritized Experience Replay （IPER） mechanism is integrated to evaluate and select actions，outputting the optimal strategy. Simulation results show that the method achieves faster and more stable decision convergence than comparative algorithms，minimizes the energy consumption for task computation offloading，and effectively adapts to changes in the number of ICVs and task sizes，demonstrating high real-time performance and excellent environmental adaptability.

An effective control strategy for fuel cell hydrogen systems can improve system dynamic performance and extend service life. In this paper，an adaptive sliding mode decoupling control strategy based on gradient optimization is proposed for circulating pump fuel cell hydrogen systems. Firstly，a fuel cell hydrogen system model is built based on Simulink. Based on this model，a decoupled sliding mode controller is designed to compensate for inaccurate model accuracy while achieving decoupling of flow and pressure. The stability of the feedback control rate is demonstrated through Lyapunov principle. However，sliding mode control has the problem of conflicting dynamic response performance and chattering. In response to this，in this study a gradient descent based sliding mode control parameter adaptive optimization method is further designed，and the system stability under variable loads is improved through a feedforward controller. At the same time，the sliding mode optimization parameter MAP self-learning method iss adopted to solve the gradient optimization delay problem under transient conditions while ensuring the stability of the closed-loop system. The results show that the adaptive sliding mode decoupling controller combined with feedforward designed in this paper has small overshoot，short response time，and high robustness. The maximum pressure difference between the anode and cathode is about 0.01 bar，and the maximum flow supply error is 0.015 g/s，which is capable of quickly responding to changes in hydrogen pressure and flow rate during variable load operation within 0.02 seconds. Compared to that before feedforward correction，the pressure fluctuation during the start-up condition of the fuel cell stack has decreased by 0.122 bar. Under disturbance，the system stability remains good，and the maximum fluctuation of hydrogen pressure is 0.01 bar.

Focusing on the demand of intelligent driving under non-visual conditions，the millimeter-wave radar with the characteristics that can work all day and is less affected by light and weather is used to build a shape and position feature recognition model of crossable obstacles on the road in this paper. Taking the road speed bump as an example，the road obstacle feature perception system based on millimeter wave radar is constructed. The radar antenna plane faces the ground and has a certain angle with the ground to collect road information. The FFT-CZT two-stage processing structure is used to refine the spectrum of radar intermediate frequency data and to obtain the range value with high accuracy. Then，by analyzing the radar point cloud，the shortest target distance measured in each frame is fused to obtain the two-dimensional imaging of the road deceleration zone. Finally，through the analysis of visual data，the geometric model of road deceleration zone is established，and the calculation method of characteristic parameters of deceleration zone is put forward. A real vehicle-testing platform is established to collect data of different angles between millimeter wave radar and the ground from 0 to 90. The average absolute error of the estimated speed bump height at the included angle of 45 is within 4 mm，and the average absolute error of the estimated width is about 21 mm，which verifies the effectiveness of the method proposed in this paper.

In order to investigate the electrical response of proton exchange membrane fuel cell （PEMFC） stack under impact load，and to reveal the mechanical-electrical coupling mechanism of PEMFC stack，the mechanical-electrical coupling modeling method of the PEMFC stack under impact load is studied. A systematic investigation is undertaken to investigate the effect of impact velocity and direction on the electrical response of the PEMFC stack，based on the established mechanical-electrical coupling model of the PEMFC stack. The results show that the proposed method for modeling the mechanical-electrical coupling of the PEMFC stack can accurately simulate the inherent mechanical-electrical coupling characteristics within the PEMFC stack. The ohmic loss of the single cell inside the PEMFC stack increases as the shock load increases. Meanwhile，the impact load results in the formation of additional electrical contact between the gas diffusion layer （GDL） and the ribs of the bipolar plate，which causes a reduction in the average value of the current density on the surface of the GDL and deterioration in the distribution uniformity. This study has certain guiding significance for the modeling of PEMFC mechanical-electrical coupling and the study of electrical response under impact load.

In addition to the design and optimization of the catalyst layer （CL），the interface between CL and the microporous layer （MPL） also needs to be considered for the research of membrane electrode assembly （MEA）. In this paper，three different CL/MPL interface structures are fabricated to verify their effect on PEMFC performance and durability under simulated vehicle operating conditions. The performance test results show that the performance of the MEA sample obtained by introducing Nafion ionomers into the CL/MPL interface （MEA-Nafion） decreases slightly compared with the pristine sample （MEA-0） at high current density，whereas the performance of the MEA sample obtained by introducing Nafion ionomers into the CL/MPL interface followed by hot pressing （MEA-Nafion-HP） is basically the same as that of MEA-0. Specially，the durability test results under simulated vehicle conditions show that the voltage decay rates of MEA-0，MEA-Nafion，and MEA-Nafion-HP samples are 42.3，29.9 and 15.2 μV/h，respectively. In conclusion，the MEA durability can be greatly improved without affecting performance by optimizing the CL/PEM interface structure design.

To deeply investigate dynamic and static mechanical properties of non-pneumatic tires under the condition of unstructured road，double-arrow cellular structure （DACS） with negative Poisson's ratio （NPR） with excellent mechanical properties is embedded into non-pneumatic tires as a novel support structure and double-arrow non-pneumatic tire （DANT） is selected as research object herein. Firstly，finite element models （FEM） of DACS with various gradient densities are developed，and quasi-static compression tests are conducted on DACS samples to verify the accuracy of FEM. Then，FEMs of DANT are established to obtain effect of layer number of double-arrow support structure on static properties of DANT，and influence of structural parameters of support structure on modal and ground properties of DANT are investigated. Finally，dynamic mechanical properties of DANT on unstructured road surfaces are investigated，including steady-state rolling，obstacle traversing，ditch crossing，and soil contact. The results show that the rolling resistance of DANT increases from 11.51 to 241.66 N with increment of radial load from 1 000 to 5 000 N. Meanwhile，a partial detachment of tread and ground come up with the increase of water velocity，and the stress distribution of support structure and the magnitude of contact stresses can be affected by the height and width of the obstacles. The spread and plastic deformation of soil occurs from contact location as soil is subjected to forces of DANT.

Autonomous commercial semi-trailers can greatly improve trunk logistics efficiency. However，since the semi-trailer vehicle is a typical underdrive system，it is difficult to simultaneously achieve lateral trajectory tracking accuracy of both tractor and trailer under large curvature motion conditions. With the increase of vehicle speed and load，the transfer of the trailer centroid and load intensifies，causing a strong uncertain impact on the hinge point between the tractor and trailer and the deviation of the driving trajectory between the trailer and the tractor to further increase，which increases the trajectory tracking difficulty and affects its driving safety. To enhance the lateral safety of semi-trailer commercial vehicle tractor and trailer，a robust path tracking strategy based on a type-2 fuzzy control algorithm is proposed in this paper. Firstly，a seven-degree-of-freedom dynamic model of the semi-trailer vehicle is constructed in MATLAB/Simulink to accurately simulate the transverse and longitudinal motion dynamics of both tractor and trailer. Secondly，considering the coverage property of the input membership function to system uncertainty in type-2 fuzzy logic control theory，a type-2 fuzzy controller is designed to adjust the lateral tracking accuracy of both tractor and trailer simultaneously. To improve the precision of lateral trajectory tracking control under uncertain factors and reduce the difficulty of controller design，a particle swarm optimization algorithm is utilized to optimize the input membership function parameters of the type-2 fuzzy controller. Finally，vehicle trajectory tracking simulation is conducted under various speed and load conditions using a joint simulation platform of MATLAB/Simulink and TruckSim to validate the control strategy proposed in this paper，and the tracking accuracy is compared with those using traditional type-1 fuzzy control and preview control. The results show that the proposed type-2 fuzzy controller can significantly enhance the lateral trajectory tracking accuracy of the tractor and trailer under the condition that track curvature changes with double shifting lines.

When driving on highways，it's necessary to reduce wind noise in the side window areas of a vehicle. Low-frequency noise control of automobile wind noise can be achieved through Active Noise Control （ANC）. Therefore，an Active Wind Noise Cancellation （AWNC） method for automobile wind noise is proposed in this paper. The suitable input signal of the side window area is selected as the reference signal，which shows good coherence with the target noise in the 100-500 Hz frequency range. Taking a full-scale clay model of the vehicle in a wind tunnel as the research object，the reference signals for wind noise are optimized through the Long Short-Term Memory （LSTM） method. The optimized reference signals are then processed using the FxLMS algorithm for AWNC simulation and validated through hardware dSPACE testing. The results show that the optimized reference signals not only reduce the number of sensors needed，thus saving cost，but also decrease the peak frequency band of wind noise by 5-15 dB.

In order to study the relationship between driver style，collision avoidance parameters and collision avoidance rate in vehicle-vehicle collision accidents to improve driving safety，in the paper five typical scenarios are analyzed by clustering videos of 610 real accidents in the Vehicle-Vehicle Traffic Accident database （VV-TRAVi，Vehicle-Vehicle Traffic Accident database with Video）. Among them，intersections account for the largest proportion，mainly divided into two typical collision scenarios with and without visual obstacles，based on the six-degree-of-freedom driving simulator to dynamically build the above two typical collision scenarios at intersections. Through questionnaire survey，principal component analysis and K-mean clustering，the driving styles are classified into aggressive，normal and cautious，with 60 volunteers of the three driving styles recruited to collect experimental data under the two typical collision scenarios constructed. The collision avoidance parameters under the three styles are extracted from the 92 sets of valid data samples： TTC （time-to-collision），BRT （braking reaction time），speed and longitudinal deceleration，and the effect of the three driving styles on the above collision avoidance parameters are analyzed using one-way ANOVA and independent samples t-test. The results show that the TTC，BRT，speed and longitudinal deceleration of the three driving styles show significant differences in the two typical collision scenarios with and without visual obstacles. Among them，the mean values （s） of the TTC for the aggressive，normal and cautious styles are 0.54 and 1.21，0.59 and 1.33，and 1.01 and 2.58，and those of the BRT are 1.12 and 0.9，1.32 and 1.3，and 1.6 and 1.56，respectively，both of which show a sequential increasing trend； the mean speed （km/h） values are 37.53 and 45.03，30.37 and 34.93，and 27.62 and 30.37，respectively，and the mean longitudinal deceleration （m/s²） values are 9.38 and 9.13，6.2 and 5.6，and 3.92 and 3.66，respectively. Both of them show a decreasing trend. The crash avoidance rates of the three driving styles are "Cautious > Normal > Aggressive". The results of the study provide a reference for the development of vehicle crash avoidance strategies that take driving styles into account.