For the problem of difficulty in balancing accuracy and real-time performance of deep learning models for intelligent cockpit driver expression recognition， an expression recognition model called EmotionNet based on attention fusion and feature enhancement network is proposed. Based on GhostNet， the model utilizes two detection branches within the feature extraction module to fuse coordinate attention and channel attention mechanisms to realize complementary attention mechanisms and all-round attention to important features. A feature enhanced neck network is established to fuse feature information of different scales. Finally， decision level fusion of feature information at different scales is achieved through the head network. In training， transfer learning and central loss function are introduced to improve the recognition accuracy of the model. In the embedded device testing experiments on the RAF-DB and KMU-FED datasets， the model achieves the recognition accuracy of 85.23% and 99.95%， respectively， with a recognition speed of 59.89 FPS. EmotionNet balances recognition accuracy and real-time performance， achieving a relatively advanced level and possessing certain applicability for intelligent cockpit expression recognition tasks.

For the uncertainty of the shift control model and parameters and the existing unknown disturbances of the two-speed dual clutch transmission （2DCT） of pure electric vehicles， a linear active disturbance rejection controller （LADRC） method is proposed. Firstly， the dynamic model of the 2DCT shifting process is established， and the shifting process is analyzed. Then， considering the uncertainty and unknown disturbance of the control model， the LADRC controller is applied to the shifting process to track the desired rotational speed， and the Extended State Observer （ESO） is used to estimate the disturbance in real time and compensate for it. Finally， it is compared with the PID control. Simulation and experimental results show that the controller has smaller tracking error and strong robustness， which ensures good gear shifting quality.

The pre-chamber ignition system can enhance the lean mixture combustion， which will then improve the thermal efficiency significantly. Based on a self-developed active pre-chamber and its fuel supply system， the effect of premixed gas injection pressure and fuel ratio on combustion characteristics at the lean boundary of an engine with a compression ratio of 16 is respectively investigated. The experimental results show that among the three kinds of premixed gas injection pressure of 0.19，0.14 and 0.09 MPa， the indicated mean effective pressure is the highest， and the ignition delay time and combustion duration are the lowest at the injection pressure of 0.09 MPa， with the most stable combustion. When the proportion of premixed gas fuel increases from 0.54% to 2.69% gradually， the engine's indicated mean effective pressure （IMEP） increases first， and then decreases， reaching its maximum at a ratio of 1.61% when the combustion is most stable and both the ignition delay period and combustion duration are the shortest. The active pre-chamber can realize the stable combustion with a λ of 1.8. Indicated thermal efficiency increases from 32.9% to 39.4%， a relatively increase of 19.8% compared with that of spark plug ignition.

Complex disturbances such as external interference， model uncertainty and parameter perturbation directly affect the accuracy and driving safety of intelligent vehicle path tracking control. Commercial vehicles are more susceptible to complex disturbances during driving because of their load characteristics. A hybrid path tracking control method is proposed in order to improve the accuracy and smoothness of commercial vehicle path tracking. Firstly， a robust sliding mode controller based on extended observer and an incremental LQR controller with stable changes are established. Particle swarm optimization algorithm is used to tune the parameters of the incremental LQR. Then， in order to improve robustness while weakening chattering， a fuzzy controller is used to adjust weight coefficient between them according to vehicle speed and lateral error. Finally， simulation analysis and vehicle experiments are conducted. The experimental data shows that SMC+LQR has good control performance to cope with complex external disturbances.

Effectively predicting the future risk indicators of multiple traffic participants under the driver's field of vision is the key to providing risk warnings to human drivers and avoiding potential collision risk. Most existing research on risk only considers the pairwise interaction between a single individual and the vehicle in the scene， and conducts research from the perspective of evaluation rather than prediction， while ignoring the different interaction between heterogeneous traffic participants and future risk status. This paper proposes a heterogeneous multi-objective risk prediction method Risk-STGCN based on spatiotemporal graph convolutional neural network， using graph convolution and temporal convolution to learn single-frame scene graph information and timing information respectively， combined with multi-layer timing prediction network to predict the multi-objective risk indicator TTC. Training and verification are conducted on the open source data set BLVD and the real vehicle self-collected data set， which is then compared with commonly used sequence prediction models. The experimental results show that the average TTC error of the proposed model on different data sets is less than 0.95 s， with multiple experimental indicators better than other models mentioned in this paper. The proposed model has good robustness and improves the interpretability of risk prediction in complex traffic scenarios.

Based on the current assessment status of the differential system of new energy vehicles， combined with the invalidation problem of the differential system installed on the cross-country type/sport sedan type， in the process of cross-country scene/racing track scene driving test， the application scenes， working principles， damage mechanisms and invalidation forms of the differential system installed on the cross-country type are analyzed in this paper. The Archard damage model is used for damage verification. Based on clustering analysis， the massive real automobile testing data collected from road test are screened and extracted， and then transformed into powertrain bench verification test conditions. The fault of the differential system of the road test automobile is reproduced， forming the reliability assessment ability of the differential system in the sand off-road scenario of new energy vehicles， which provides support for the selection and optimization of differential systems and the development of new automobile models. Meanwhile， this article also has certain reference value on how to improve the performance and reliability of differential systems.

A trajectory planning method based on graph search and optimization is proposed for intelligent vehicle trajectory planning in dynamic unstructured environments. Firstly， the graph search method is employed to search for motion primitives for intelligent vehicles to obtain initial trajectories that conform to kinematic characteristics. Then， based on nonlinear model predictive control methods， the trajectory is optimized to obtain smoother and safer trajectories. In order to achieve rapid and secure expansion of primitives in dynamic unstructured environments， a method for primitive collision detection is proposed. This method uses obstacle expansion and grid discrete motion elements to perform static collision detection on irregular obstacles， and introduces in the concept of velocity obstacles to perform dynamic collision detection on dynamic obstacles in velocity space. The proposed algorithm is compared by simulations in ROS/Gazebo environment， and is evaluated by field tests. The results show that compared to the TEB algorithm， the proposed trajectory planning method improves the average obstacle avoidance success rate by 18% while meeting the real-time computing requirements， demonstrating higher safety obstacle avoidance ability and feasibility.

The evaluation of scenario complexity is crucial for improving adaptability and flexibility of autonomous vehicles in coping with complex environments and enhancing the applicability of the algorithms. A graph-based algorithm for evaluating scenario complexity is developed in this paper， which fully considers interactive topology and categorizes traffic scenarios into three complexity levels. The reasonability and effectiveness are validated in ramp merging scenarios. To demonstrate its scalability， the evaluation algorithm is applied in the development of the trajectory prediction and decision-making algorithms of automated driving. The proposed algorithms are then tested using natural driving datasets and vehicle-in-the-loop experiments. The results indicate that scenario complexity evaluation enables early estimation of prediction uncertainty， enhances the real-time and optimality of decision-making algorithms. In data replay tests， the complexity assessment module can reduce the failure rate and collision rate during lane merging by approximately 38% and 92%， respectively， indicating promising application prospects.

Air conditioning system as a key subsystem of environmental regulation within the entire vehicle system， the carbon emission of air conditioning system throughout its entire lifecycle is crucial for meeting the environmental protection and emission requirement of electric vehicle. Combining the life cycle climate performance （LCCP） model of electric vehicle air conditioning system with relevant data， the LCCP values in different provinces of China are analyzed in this paper. Besides， the LCCP values under two different heating schemes and different carbon intensities of electricity are compared. The results show that low-GWP mixed refrigerant RE170/R134a （RE170 to R134a mass fraction ratio 90∶10） can lead to a decrease in LCCP values for electric vehicle air conditioning system by 11.2% to 28.1% in China. Replacing the PTC heater with the heat pump results in a decrease in LCCP values by 0 to 33.1%. In addition， considering the future changes in China's carbon intensity of electricity and the proliferation of electric vehicles， it is anticipated that the LCCP values for an individual vehicle by 2035 will decrease by 31.7% to 39.3%， while the gross electric vehicle LCCP values in China will increase significantly.

For the stability control problem of distributed four-wheel-drive electric vehicles under extreme conditions， considering the influence of sensor noise of yaw rate， lateral and longitudinal acceleration， as well as the estimation error of slip angle， a phase plane stability domain division method based on extension theory and an adaptive Tube-based Model Predictive Control algorithm （ATMPC） are proposed to quickly quantify the stability level of the vehicle and ensure the vehicle driving stability while maintaining tracking accuracy. The designed vehicle yaw stability control system utilizes hierarchical design architecture. The upper layer employs the extension theory to associate the vehicle slip angle-yaw rate phase plane with extension control domain and determines the control domain based on the actual vehicle state and calculates the dependent function to realize the decision-making of the control target weights and modes of the lower layer's Tube-MPC. The lower layer utilizes Tube-MPC to track the desired vehicle slip angle and yaw rate， enabling precise decision-making regarding the yaw moment， and adopts the tire loading ratios optimization method for the allocation of the yaw moment. The control strategy is validated by Carsim/Simulink co-simulation. The results show that the proposed control framework and ATMPC strategy can significantly enhance the driving stability of vehicles in extreme conditions and improve robustness in noisy environments， outperforming traditional MPC.