In order to improve the fuel economy of Plug-in Hybrid Electric Vehicle (PHEV), this paper proposed the vehicle speed prediction method based on real-time traffic information. The energy management strategy was established to obtain the optimal fuel economy based on Model Prediction Control (MPC), the real-time optimal torque distribution was optimized with the help of dynamic programming algorithm. Simulation verification was conducted on MATLAB/Simulink platforms which showed that the accuracy was improved by 13.5% compared with traditional vehicle speed prediction methods in real-time road conditions. Compared with the MPC strategy based on historical vehicle speed, the fuel economy of MPC strategy based on real-time traffic information is improved by 9.5%.

A hierarchical control strategy based on Radial Basis Function Neural Network (RBFNN) and Stochastic Dynamic Programming (SDP) was proposed for Plug-in Hybrid Electric Vehicle (PHEV) queue. Firstly, the powertrain structure and mathematical model of PHEV were analyzed in detail, then a hierarchical control framework was constructed. The upper layer adopted RBFNN to train driving data derived from Model Predictive Control (MPC) to generate the speed tracking controller. According to the information of the speed and power demand transmitted from the upper layer, a Markov chain model was established for the lower layer controller, the Markov chain model can realize the optimal energy distribution between PHEV traction battery and the engine based on the SDP theory. The simulation results show that compared with CD/CS strategy and rule-based strategy, the energy consumption of PHEVs in the queue is significantly reduced while ensuring safe driving under high-speed conditions based on the proposed strategy.

For the problems of strong randomness and low training efficiency in intelligent vehicle training under reinforcement learning algorithm, this paper proposed a driving decision framework of intelligent vehicle based on rule constraints and Deep Q Network (DQN) algorithm. The introduced rules were divided into hard constraints related to lane change and soft constraints related to lane keeping, which were implemented by Action Detection Module and reward function respectively. At the same time, the network structure of DQN was improved by combining Dueling DQN and Double DQN, N-Step Bootstrapping learning was introduced to accelerate the training efficiency of DQN. Finally, the effectiveness of the model was verified by comprehensive comparison with the original DQN algorithm in the highway scene of Highway-env platform. The improved algorithm improved the task success rate and training efficiency of intelligent vehicles.

For the energy management of new energy vehicles, it is difficult to predict the vehicle speed in a long-term and accurate way, this paper proposed a model-based parametric prediction method to predict the vehicle speed trajectory using the forward-looking data provided by sensors and GPS. Firstly, the speed prediction algorithm based on Intelligent Driver Model (IDM) was established in term of vehicle dynamics and vehicle stop-turn trend; secondly, data was selected from the NGSIM public data set for parameter calibration and simulation; then the algorithm parameters were calibrated using Genetic Algorithm (GA). The results show that the optimized speed prediction algorithm has high accuracy for long-term speed prediction in both unobstructed and congested traffic environments, the error can be controlled in the range of 8%~13%.

To improve the performance of vehicle detection algorithm based on 3D point clouds, this paper proposed a real-time vehicle target detection algorithm based on the bird’s-eye view of point cloud. First, the original 3D vehicle point cloud was converted into the 2D point cloud RGB feature map. Second, the vehicle’s yaw angle prediction branch was added to achieve accurate vehicle localization based on YOLOv4-tiny network, the target localization capability of the network was improved by adding an improved Spatial Pyramid Pooling-Fast (SPPF) module. Finally, the target detection precision was improved by introducing a dual-attention mechanism and optimizing the loss function in the backbone network. The test results show that the average vehicle detection precision of the proposed algorithm reaches 96.92%, which is 2.94 percentage points higher than YOLOv4-tiny, the average detection accuracy of the proposed algorithm reaches 87.73% in the moderate difficulty of KITTI bird’s-eye view validation set, detection rate reaches 100 frames per second, which is able to meet the real-time requirements.

Generative Artificial Intelligence (Generative AI), a technology based on neural network models for content generation, has received widespread attention in the industry and academia in recent years. With the continuous deepening of the application of this technology in various fields, the large model based on Generative AI has also a huge impact on the transformation of technical solutions in the field of autonomous driving. This article provided a brief overview of the development of Generative AI technology and large models, including their classification methods and representative models. At the same time, this article also delved into the application of generative models for the field of autonomous driving. Finally, a summary and prospect were conducted in the future development direction of Generative AI technology and autonomous driving technology.

In order to accurately evaluate the dark current of pure electric bus and avoid vehicle failure to start due to insufficient battery power, 2 models of pure electric buses were selected, the power supply mode and vehicle dark current of their on-board electrical equipment were analyzed systematically. At the same time, on the basis of summarizing the conventional dark current measurement methods, the long-scale measurement method was proposed, using the electric power tester and measurement module to build a detection scheme to record the dark current and complete the real vehicle test. The results of theoretical analysis and experimental data comparison show that the method based on long scale measurement can accurately evaluate the dark current value of pure electric bus and control it in a reasonable range, then guide the calculation of vehicle outage time.

In order to improve vehicle aerodynamic coefficients comprehensively, this paper proposed a shape optimization design scheme. Firstly, vehicle without crosswind was simulated numerically by using realizable k-ε turbulence model. The reliability of the simulation model was verified by wind tunnel tests. On this basis, the influence of different crosswind angles on the aerodynamic characteristics was studied, the aerodynamic coefficients of yaw angle of 12° were taken as the reference benchmark for optimization, samples were extracted by uniform Latin hypercube for flow field calculation, the response surface model was used to approximate the corresponding relationship between automobile modeling parameters and aerodynamic coefficients, the Pareto front solutions were obtained based on the genetic optimization algorithm. Finally, 4 optimization schemes were determined, which reduced the drag coefficient by 2.6%, the lateral force coefficient by 6.54%, and the lift coefficient tends to be negative, effectively improving the aerodynamic characteristics of the vehicle.