In order to solve the shortcomings of traditional object detection and tracking algorithms, such as low detection accuracy, poor global perception ability, poor recognition ability of occlusion and small target objects, this paper proposed a vehicle tracking method based on YOLOv5 and DeepSORT algorithm improved by lightweight Transformer. Firstly, the EfficientFormerV2 model was used to improve the YOLOv5 algorithm model to enhance the target detection ability of the vehicle, and then the advantages of the Swin model were used to improve the Re-Identification module in the DeepSORT multi-target tracking algorithm to enhance the tracking ability and accuracy of the vehicle. Finally, the dataset KITTI and VeRi were used to carry out comparative experiments and ablation experiments. The results show that under complex conditions, the performance of the proposed method is significantly improved in vehicle occlusion and small target recognition, with an average accuracy of 96.7%, an increase of 9.547% in target tracking, and a reduction of 26.4% in the total number of ID switching.

There are insufficient vehicle-vehicle interaction and poor matching between planning and control in the decision-making model of high-speed autonomous vehicles. In order to solve these problems, a closed-loop lane change decision model based on Stackelberg game was constructed. Faulty vehicle response was incorporated into lane changing decision while introducing driving style feature. The multi-objective decision-making cost function was optimized. Particle Swam Optimization (PSO) algorithm was used to solve the game decision model, and the vehicle state was predicted by using a kinematic model that considers the influence of center of mass sideslip angle. A nonlinear model predictive planning controller based on dynamic risk potential field method was designed. The simulation test results show that the closed-loop lane-changing decision-making model proposed in this paper can effectively combine the interaction behavior and driving style characteristics of vehicles to make correct decision-making instructions and implement corresponding motion planning and control.

In view of the strong coupling of each control factor in the lateral control of autonomous vehicles, it is difficult for the control method relying on the ideal model to completely decouple and migrate from the simulation environment to the actual vehicle, and the problem that the convergence speed of the reinforcement learning method in the lateral control of autonomous vehicles is not ideal, the fuzzy inference machine and the similarity of the simulation reinforcement learning in the lateral control factors of vehicles are used to combine the two. A fuzzy inference machine is used as the initialization condition for simulated reinforcement learning, and provide guidance for the learning process, thus achieving rapid convergence of the learning process. The MATLAB/Carla simulation and vehicle test are applied to verify the control method. The results show that the method can significantly reduce the number of simulation reinforcement learning iterations, achieve better vehicle lateral control performance in 500 full path iterations, and achieve good control effect in both simulation and real environment on the basis of not relying on the ideal mathematical model and not having to carry out in-depth optimization of the fuzzy inference device.

As the battery temperature shows a high distribution in the middle and low distribution around during the operation of electric vehicles, it affects the battery service life. To this end, a steady-state heat conduction structure design method based on Floating Projection Topology Optimization (FPTO) is proposed using the geometric mean temperature as the objective function, and the maximum temperature and temperature difference of the power battery pack under single and multiple operating conditions are comparatively analyzed by means of an arithmetic example, which demonstrates that the obtained topological configuration can effectively reduce the maximum temperature and temperature difference in the heat dissipation process during the multiple operating conditions, so as to make the temperature distribution uniform. Finally, the method is applied to the power battery pack support structure, and the results show that the method effectively reduces the temperature without increasing the volume of the battery pack structure, optimizes the distribution of materials, and realizes lightweight of the overall structure of the battery pack.

In order to analyze the contribution of different frequency bands of noise to the annoyance degree of a low-speed and high-torque electric drive assembly, a low-speed and high-torque electric drive assembly was used as the research object, combined with psychoacoustic theory to establish a tone theory analysis model of the electric drive assembly. The difference in the degree of the influence of the tone on the subjective annoyance at different frequencies of the electric drive assembly was considered, and the theoretical analysis model of the modified tone (annoyance) was established. Combined with the noise source distribution characteristics of the electric drive assembly, the low frequency noise (0~400 Hz), intermediate frequency noise (>400~2 500 Hz) and high frequency noise (>2 500 Hz) caused by the electric drive assembly were compared and analyzed. The results show that the most important factors affecting the annoyance level are electromagnetic force wave and switching frequency induced noise, followed by transmission gear noise.

To study the applicability of different data correlation evaluation methods for virtual evaluation of far side occupant protection condition, firstly, the dynamic response of 11 simulation and test data was determined by analyzing far side occupant basic data of two vehicle models, CORrelation and Analysis (CORA) and ISO/TS 18571 methods were further applied for correlation evaluation, and error analysis method was used to analyze the response with significant differences in the correlation evaluation results, and it is concluded that ISO/TS 18571 method provides more reasonable validation results than CORA method.

To address the issue of front brake low-frequency squeal of passenger car, a mathematical model and multi-condition matrix of stability analysis of braking system were established, and stability of the braking system was analyzed with complex mode simulation. The results show that the unstable mode appears in the system under 1 830 Hz, which is in good agreement with the vehicle brake squeal test results. Components and their modal contribution analysis method were used to analyze the unstable mode. The results show that the first torsional mode of the caliper and the first bending mode of the pad are the biggest contributors to the unstable mode. The brake caliper and pad were optimized. The results show that the 1 830 Hz unstable mode becomes stable after the caliper structure is strengthened, the system stability is improved after the pad material and structure adjustment, whereas it is still in the unstable interval, and the squeal is difficult to remove, which is consistent with the experimental results, and the subjective evaluation result is acceptable.

In order to reduce load fluctuations and network losses caused by large-scale electric vehicles connected to the distribution network, this paper proposed a multi-objective optimization control method based on Particle Swarm Optimization (PSO) algorithm for flexible loads of large-scale electric vehicle charging connected to the distribution network. Firstly, a coupling model between transportation network and distribution network was established, and combine it with the travel chain model to analyze users’ charging needs, and a prediction model for the energy state of connected electric vehicles was established; Secondly, the minimized standard deviation of load fluctuations and network losses in the distribution network was taken as the optimization objective, and a multi-objective optimization function was established for the flexible load integration of large-scale charging of electric vehicles into the distribution network, meanwhile distribution entropy was introduced to design inertia weight update strategy and optimize PSO algorithm. Finally, the improved PSO algorithm was used to achieve flexible load control of the distribution network based on functional constraints. The test results show that the proposed method can accurately analyze the charging needs of users, and reduce the peak load fluctuation and network loss of the controlled distribution network.

To improve the prediction accuracy of electric vehicle charging load, a multi time scale prediction model for electric vehicle charging load was designed based on the Lightweight Gradient Boosting Machine (LightGBM) algorithm and travel chain theory. The travel chain was used to describe the user’s travel process, Monte Carlo method was used to extract the spatiotemporal data, and the probability density functions of travel and stay time in different regions was calculated. Newton method was used to divide the probability of charging at multiple time scales, clarifying the spatiotemporal distribution of driving and charging conditions. Fuzzy mathematics theorem and LightGBM were applied to classify charging load data, and a multi season and multi time prediction model were constructed. The efficient parallel computing mode of LightGBM was applied which clarified the variation pattern of charging load, and multi time scale prediction was achieved. The experimental results show that the established model has a prediction error of less than 100 kW and a prediction false alarm rate of less than 3% under different seasons and the number of electric vehicles, and can accurately display the variation pattern of charging load.

In order to accurately predict the V2G charging and discharging load of electric vehicles, so as to regulate the peak to valley difference of power grid load and ensure power supply stability, this paper proposed a spatiotemporal distribution prediction method for V2G charging and discharging loads of electric vehicles based on collaborative optimization of supply and demand sides. A collaborative optimization objective model for both supply and demand sides was built, the Whale Optimization Algorithm was used for iterative solution to obtain the optimal charging and discharging load curve, and the optimal charging and discharging period was determined. The influencing indicators of charging and discharging loads within the optimal time periods in different spatial regions were collected, serving as inputs for constructing a prediction model based on multiple linear regression, thus achieving the prediction of spatial-temporal distribution of electric vehicle V2G charging and discharging loads. The experimental results show that the load prediction model obtained with the proposed method has a relatively large coefficient of determination, indicating that the prediction results of this research method are closer to the actual load, and have high prediction accuracy.