In recent years， with the continuous enhancement of fuel cell power， hydrogen supply systems have been evolving towards blind-end anode topologies with hydrogen circulation. However， research on testing systems for hydrogen supply and circulation lags noticeably， particularly in the performance testing of core components such as hydrogen circulation pumps and injectors. Therefore， a multifunctional testing platform for fuel cell hydrogen supply systems is developed in this paper， enabling component testing， characteristic data acquisition， offline calibration， and other functionalities for hydrogen circulation systems with different configurations. The platform， by simulating the pressure drop， hydrogen consumption， and the production of water and heat in real fuel cells， mitigates the additional cost incurred by testing on the performance and lifespan of actual fuel cells. Ultimately， based on this platform， an anode pressure control and anode purge control test are conducted on the hydrogen supply system of a 150 kW fuel cell， verifying the capability of the developed testing platform to meet specific testing requirements for different loads.

To ensure the reliability of aluminum-plastic film encapsulation for lithium-ion batteries， strict control of the aluminum layer thickness after forming is necessary. However， obtaining the thickness relies heavily on physical experiments， resulting in high cost for both early design optimization and later production process quality monitoring. In this paper， a combination of physical experiments and simulation modeling is adopted to establish a constitutive equation that can well characterize the mechanical properties of the pouch during forming. Additionally， a prediction method for the aluminum layer thickness based on the overall aluminum-plastic film thickness is proposed， enabling precise prediction of the aluminum-plastic film and aluminum layer thickness after forming. Furthermore， based on simulation Design of Experiments （DOE）， key influencing factors are screened to construct a response surface model， facilitating rapid prediction and optimal parameter matching design for different products， which also provides a solution for online monitoring of forming quality during production. The results show that the multi-layer composite aluminum-plastic film exhibits obvious anisotropy during the plastic stage. The 3-parameters Barlat-Lian constitutive model effectively represents the anisotropic properties of the film， and outperforms the single-directional elastic-plastic model， achieving accurate prediction of the aluminum-plastic film performance after forming. The constructed response surface model can replace the refined finite element model， and have excellent prediction accuracy for the thickness of the composite aluminum-plastic film and the aluminum layer， with an error less than 5%. By optimizing the process parameters， the formed thickness of the aluminum layer can be increased by 10%~20%. The integrated development application APP can meet the requirements for quick design evaluation， parameters optimization， and online monitoring of the forming quality.

There may be inconsistencies in temperature， charge state， aging state （capacity and internal resistance） between individual cells in a battery module. Due to the existence of the "short board effect"， the inconsistencies will affect the overall performance of the battery module， so timely and accurate inconsistencies diagnosis is very necessary. Considering that the above-mentioned inconsistencies will affect the electrode process characteristics， which will be reflected in the Electrochemical Impedance Spectroscopy （EIS） and Distribution of Relaxation Time （DRT）， in this paper， after clarifying the effect of several kinds of inconsistencies on EIS and DRT by combining the equivalent circuits， an inconsistencies diagnosis method for battery modules based on EIS and DRT is innovatively proposed. The performance of unsupervised clustering algorithms such as K-means， AP （Affinity Propagation） and DBSCAN （Density Based Spatial Clustering of Applications with Noise） is comparatively analyzed by mixing the abnormal batteries into a group of batteries with good consistency. The results show that the DBSCAN diagnostic accuracy is 99.2%， which can realize the accurate diagnosis of the inconsistency difference of single cells within the battery module.

Accurate performance evaluation of power battery cells is of great significance to ensuring the safety of power batteries. For the existing data-driven battery fault diagnosis algorithms， mostly individual cells are compared with each other and the outlier cells are identified as faulty cells by classification， based on differences in characteristic parameters such as single cell voltage. However， if there are multiple cells of similar abnormally performance in the power battery pack， or all individual batteries show an overall performance deterioration， it is difficult to distinguish individual cells or even there is no significant outliers， and the application of the mutual comparison strategy is limited. A power battery fault diagnosis method is proposed based on 1dCNN-LSTM to quantify the abnormality of a single cell in this paper. Combining the three types of characteristics of vehicle motion status， drive system status and power battery electrical signal， the 1dCNN-LSTM fusion model is established to estimate the individual cell voltage under ideal conditions as reference. The difference between the real-time voltage reference value and the measured voltage value is used to quantify the abnormality of each cell. Combined with actual cases， it is shown that for thermal runaway case due to single cell failure， the abnormal performance of the faulty cell compared to others can be identified 7 days before accident， and potential risk can be recognized in discharge processes from a year of more before the accident. For overall deterioration cases without obvious individual cells inconsistency， the deterioration evolution within the last 7 days can be tracked.

A lithium battery early fault diagnosis method based on WOA-VMD and Shannon entropy is proposed in this paper to solve the problem of current battery management systems being unable to diagnose early faults. Firstly， the whale optimization algorithm is introduced to optimize the parameters of the variational mode decomposition algorithm to improve its decomposition performance and obtain intrinsic mode function components containing more fault feature information. Then， the voltage signal of the individual battery is decomposed and reconstructed to reduce the impact of measurement noise and additional excitation voltage. Furthermore， a sliding window is used to calculate the Shannon entropy range of individual voltage and the overall Shannon entropy of individual voltage dispersion to set appropriate thresholds for early fault diagnosis. After verification with actual vehicle data， this method can provide fault warning about 10 minutes in advance without generating false warnings for vehicles without faults. It has strong robustness and reliability.

For the trade-off between prediction model accuracy and computational cost for path tracking control of autonomous vehicles， a learning-based model predictive control （LB-MPC） path tracking control strategy is proposed in this paper. A two-degree-of-freedom single-track vehicle dynamic model is established， and an in-depth analysis is conducted on its step response error with respect to variation in vehicle speed， pedal position， and front wheel steering angle compared to the IPG TruckMaker model. Methods for constructing error datasets and receding horizon updates are designed， and the Gaussian process regression （GPR） is employed to establish an error-fitting model for real-time error compensation and correction of the nominal single-track model. The error correction model is utilized as the prediction model， and a path tracking cost function is designed to formulate a quadratic programming optimization problem， proposing a learning-based model predictive path tracking control architecture. Through joint simulation using the IPG TruckMaker & Simulink platform and real vehicle experiments， the real-time performance and effectiveness of the proposed GPR error correction model and LB-MPC path tracking control strategy are verified. The results show that compared to the traditional model predictive control （MPC） path tracking control strategy， the proposed LB-MPC strategy reduces the average path tracking error by 23.64%.

The three-dimensional object detection algorithm based on point cloud is one of the key technologies in the autonomous driving system. Currently， the voxel-based anchor-free detection algorithm is a research hotspot in academia， but most researches focus on designing complex refinement stage， at the expense of huge algorithm latency， to bring limited performance improvement. Although the single-stage anchor-free point cloud detection algorithm has a more streamlined detection process， its detection performance cannot satisfy the needs of autonomous driving scenarios. In this regard， based on the anchor-free detection algorithm CenterPoint， a single-stage anchor-free point cloud object detection algorithm for autonomous driving scenarios is proposed in this paper. Specifically， the encoding and decoding sparse module is introduced in this paper， which greatly promotes the information interaction of the spatial non-connected areas of the three-dimensional feature extractor， ensuring that the three-dimensional feature extractor can extract features that satisfy various target detection. In addition， considering that it is challenging to adapt the existing two-dimensional feature fusion backbone to the center-based head， in this paper self-calibrated convolution and large kernel attention modules are introduced in to effectively extract point cloud features of the target area， which are then gathered into the center point area， thereby improving the algorithm's recall and accuracy of the target. The proposed algorithm in this article is trained and experimentally verified on the large-scale public dataset of nuScenes. Compared with the benchmark algorithm， mAP and NDS are increased by 5.97% and 3.62% respectively. At the same time， the actual road experiments with the proposed algorithm are conducted on a self-built vehicle platform， further proving the effectiveness of the proposed algorithm.

To address limitation in prediction accuracy and data utilization efficiency of supervised learning-based trajectory prediction models， a trajectory prediction model and a general self-supervised pretraining strategy are proposed. Firstly， a lightweight trajectory prediction model based on Transformer is established to extract temporal-spatial features while modeling interaction relationship. Secondly， three types of masks， namely motion information temporal mask， road information spatial mask， and interaction relationship mask， are designed for self-supervised pre-training tasks on the model to enhance the model's ability to extract general scene features. Finally， pretraining weights are used as initialization parameters for supervised learning fine-tuning in downstream tasks. Experimental results on the Argoverse2 Motion Forecasting dataset show that the model can effectively reconstruct traffic scenes in pretraining tasks. The introduction of self-supervised pretraining improves prediction accuracy and data utilization efficiency. Moreover， it exhibits universality for different prediction tasks， achieving a 3.3% and 3.7% improvement in the minFDE₆ for single-agent and multi-agent trajectory prediction tasks， respectively.

Multi-sensor fusion is an effective way to improve intelligent vehicle perception. For the data-matching problem of the three types of sensors of LiDAR， millimeter-wave radar， and camera， traditional methods such as bipartite graph matching can’t achieve high precision， with poor matching robustness. Therefore， a multi-sensor data fusion algorithm for intelligent vehicles based on tripartite graph matching is proposed in this paper. The problem of data matching of the three sensors is abstracted as a weighted tripartite graph-matching problem. By using Lagrange relaxation， the original problem space is decomposed into subspaces， the weights of vertices and edge inside which are determined then by the cost matrix model. Furthermore， combining the perceptual error model and likelihood estimation， the posterior distribution of perceptual errors is determined. Ultimately the Lagrange Multiplier （LM） model is used for data matching. Finally， the effectiveness of the proposed matching algorithm is validated by the nuScenes training dataset and real-world vehicle tests. On the dataset， the proposed algorithm improves F1 scores by 7.2% compared to common algorithms. In various real-world vehicle scenarios， the proposed algorithm shows excellent perceptual accuracy and robustness across.

The development of visual perception technology based on deep learning is beneficial for the advancement of environment perception technology in automatic driving systems. However， for corner cases of autonomous driving scenario， there are still some problems in the current perception model. This is because the ability of the perception model based on deep learning depends on the distribution of the training dataset. Especially when categories in the driving scene never appear in the training set， the perception system is often fragile. Therefore， identifying unknown categories and extreme scenarios remains a challenge for the safety of automatic driving perception technology. From the perspective of processing data sets， in this paper a novel multimodal automatic corner case mining process called "Corner Case Mining Pipeline （CCMP）" is proposed. In order to verify the effectiveness of "CCMP"， the concern case subset "Waymo-Anomaly" on the basis of Waymo open datasets is established， with a total of 3 200 images， each of which will contain the corner case scene defined in the text. Then based on the private data set Waymo-Anomaly， it is proved that the recall rate of "CCMP" corner case mining can reach 91.7%. In addition， the effectiveness of object detectors targeting long-tailed distributions in datasets containing corner case is experimentally verified. Ultimately， the authenticity of the automatic driving perception model in the real world is expected to improve from the perspective of datasets processing.