To address the challenges in the communication between infrastructure and vehicles in intelligent transportation systems, this paper applies the Intelligent Reflecting Surface (IRS) to the intelligent transportation system based on visible light transmission, and deeply analyzes its channel transmission model and the influence of road infrastructure parameters. By modeling the Non-Line-of-Sight (NLOS) channel between the infrastructure and vehicles,the transmission performance of signals in the NLOS channel is analyzed. The closed-form analytical solutions of the system performance are derived for the bit error rate, downtime probability, and block error rate of the IRS-assisted NLOS communication. The impact of the compound pointing error on the bit error rate, downtime probability, and block error rate of the IRS-assisted NLOS communication system is analyzed. The bit error rate performance under M-ary Phase Shift Key (MPSK) modulation and traditional key control modulation is also analyzed. The experimental results show that under the IRS-assisted NLOS transmission condition, when the link follows the double-Rayleigh distribution or the Rayleigh-Rice combined distribution, increasing the distance between the transceiver devices and the height of the street lamp will increase the bit error rate. Increasing the downtime threshold can increase the downtime probability. Increasing the number of transmitted bit blocks can improve the block error rate. MPSK modulation can reduce the system bit error rate more rapidly, and the performance is enhanced as the modulation order increases. After considering the pointing error to the system, its bit error rate, downtime probability and block error rate all increase. The proposed scheme can effectively enhance system communication performance.

To mitigate the interference of external disturbances and environmental uncertainties and improve vehicle speed tracking accuracy, this study proposes a longitudinal motion controller that integrates a High Gain Extended State Observer (HGESO) with Model Predictive Control (MPC). First, the multi-source external uncertainties are consolidated into a stochastic time-varying resistance term in the speed control framework, which is estimated using the HGESO. This approach is combined with a nominal state-space model to enhance the description of vehicle longitudinal dynamics. Subsequently, an incremental MPC controller incorporating the estimated disturbance resistance is employed. This controller designs a multi-objective optimization function that simultaneously considers longitudinal speed tracking error, ride comfort, and energy consumption, ultimately solving for the optimal control input. Finally, precise calibration of the lower-level controller’s mapping table is performed to ensure accurate output of throttle and brake commands, thereby enhancing the controller’s real-time execution capability. Experimental results demonstrate significant improvements in speed tracking accuracy under challenging conditions: a 35%~61.54% enhancement is achieved on steep slopes, and a 26.3%~80.8% improvement is observed during continuous steering maneuvers. The proposed control strategy effectively eliminates the impact of external disturbances on vehicle longitudinal control.

To achieve accurate recognition of vehicle driving intentions in highway scenarios, this paper proposes a driving intention recognition model that combines dual reference lines in the Frenet coordinate with Gaussian Mixture Models (GMMs) and Hidden Markov Models (HMMs). The model selects driving data from different reference lines in the Frenet coordinate based on vehicle position as observed variables. By integrating the observation probabilities output by the GMM at previous and subsequent time steps with the HMM, the model identifies the vehicles’ driving intention at the current moment. The effectiveness of the model is validated using the US-101 dataset from NGSIM. The results show that the dual-reference-line GMM-HMM model achieves recognition accuracies of 93.33% for lane keeping and 92.24% for lane changing, indicating excellent recognition performance.

For the problem of mapping failure in the high-speed tracking and figure-eight scenarios of the Formula Student Autonomous China (FSAC) due to the limited recognition and low accuracy of single-sensor cone detection, this paper proposes a cone mapping algorithm based on the loose coupling of LiDAR, industrial cameras, and a combined inertial navigation system. By projecting LiDAR data onto the camera coordinate system, the similarity between the target detection bounding boxes from the camera’s deep learning framework (YOLOv5) and the LiDAR cone bounding boxes is matched. The fused point cloud, containing RGB color information, is then transformed from the LiDAR coordinate system to the map coordinate system. The real-time vehicle pose calculated by the combined inertial navigation system is used to update the fused cone point cloud map. Real-vehicle comparative test results show that the algorithm achieves an average recall rate of 98.6% and an average precision of 99.1%, enabling the distinction between the inner and outer tracks of the cone map, thereby enhancing the vehicle’s perception, anticipation capabilities and path planning efficiency.

In view of the exponential increase in the number of key scene scenarios generated in high-dimensional space, and the difficulty of traditional artificial construction or random search methods to balance coverage and efficiency, this paper proposes a search method based on single-objective Tree structure Parzen Estimator (TPE) and Multi-ObjectiveTree structure Parzen Estimator (MOTPE). A software-in-the-loop automated simulation testing framework is built by using the CARLA simulator. Taking weather elements as an example, the critical scenario generation effects of the different search algorithms are compared. The experimental results indicate that the TPE-based search method and the MOTPE-based method increase the number of key scenarios generated by 3.11 times and 2.06 times, respectively, compared to the random search method. The MOTPE method is 1.53 times better than TPE in terms of scenario quality. When combined with scenario automaed generation and testing frameworks, these methods effectively address the issue of exploding scenario numbers, allowing for the discovery of scenarios with high testing value.

Aiming at the issue of lacking slope information in urban driving cycles used for vehicle performance evaluation, this paper proposes a method for Urban Ramp Driving Cycle (URDC) construction based on Self-Organizing Map (SOM) neural network. Typical road driving data with urban ramp characteristics is collected using the average traffic flow method. After pre-processing, the data is segmented into short trips, and 20 parameters representing road operation characteristics are selected as the feature parameters of the short trips. The dimensionality of these feature parameters is then reduced via principal component analysis, followed by clustering the short trips analysis using a SOM neural network. According to the principle of smooth ramp connection, short trips with high correlation are selected to construct an urban ramp driving cycle that includes both speed and slope information. The results of automatic transmission operated in slope performance test indicate that the constructed driving cycle can reflect the driving characteristics of vehicles on road with urban ramp features, which can be used as the benchmark driving cycle for performance test of vehicle driving on urban ramps.

This paper studies the noise of a Permanent Magnet Synchronous Motor (PMSM) for electric vehicles. Through the analysis of the contribution of different electromagnetic excitations to the noise, the paper identifies key factors such as motor torque ripple and spatial zero-order electromagnetic force, as well as their contributions to the order noise in different speed ranges. Moreover, through the simulation analysis of the noise under the conditions of eccentricity and current harmonics, the changes in the order noise after considering practical factors are obtained. The Kriging method is used to create an electromagnetic surrogate model, which can quickly calculate the electromagnetic force and torque ripple based on the design parameters of the stator and rotor. On the basis of the surrogate model, the genetic algorithm is used for multi-objective optimization. The order noise obtained from the simulation is significantly reduced compared with the original scheme. The good consistency between the simulation and the experiment is verified through the noise test of the optimized prototype on the test bench.