Yan Sun

The cathode and anode differential pressure control of a proton exchange membrane fuel cell (PEMFC) directly affects its service life and operating efficiency. Existing control methods find it difficult to cope with strong nonlinear perturbations, and fixed differential pressure control is prone to pressure overshoot and threshold exceedance, resulting in unstable pressure regulation. In order to solve the current research problems, a reinforcement learning method based on hybrid experience replay (HP-TD3) is proposed. A CART-based algorithm is first used to classify the states of the test load, and a load-related segmented reward function is designed. In addition, a hindsight experience replay (HER) mechanism is incorporated into the Priority Experience Replay Twin Delayed Deep Deterministic Policy Gradient (PER-TD3) framework to improve sample utilization efficiency and training stability. Finally, the performance of HP-TD3 and its ability to cope with nonlinear disturbances are verified on a fuel cell control unit hardware-in-the-loop (FCU-HIL) platform. In the A test load (frequent switching and high low-load proportion), the Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and the degradation index of the fuel cell dynamic performance (Δfc) of HP-TD3 are respectively reduced by 17.4%, 20.5%, and 13.3% compared to P-TD3; in the B test load (high-load operation and low switching frequency), these indicators are reduced by 25.7%, 29.4%, and 15.4% respectively.

Recent breakthroughs in artificial intelligence are accelerating the intelligent transformation of vehicles. Vehicle electronic and electrical architectures are converging toward centralized domain controllers. Deep learning, reinforcement learning, and deep reinforcement learning now form the core technologies of domain control. This review surveys advances in deep reinforcement learning in four vehicle domains: intelligent driving, powertrain, chassis, and cockpit. It identifies the main tasks and active research fronts in each domain. In intelligent driving, deep reinforcement learning handles object detection, object tracking, vehicle localization, trajectory prediction, and decision making. In the powertrain domain, it improves power regulation, energy management, and thermal management. In the chassis domain, it enables precise steering, braking, and suspension control. In the cockpit domain, it supports occupant monitoring, comfort regulation, and human–machine interaction. The review then synthesizes research on cross-domain fusion. It identifies transfer learning as a crucial method to address scarce training data and poor generalization. These limits still hinder large-scale deployment of deep reinforcement learning in intelligent electric vehicle domain control. The review closes with future directions: rigorous safety assurance, real-time implementation, and scalable on-board learning. It offers a roadmap for the continued evolution of deep-reinforcement-learning-based vehicle domain control technology.

Hydrogen has the physical and chemical characteristics of being flammable, explosive and prone to leakage, and its safety is the main issue faced by the promotion of hydrogen as an energy source. The most common scene in vehicle application is the longitudinal wind generated by driving, and the original position of hydrogen concentration sensors (HCSs) did not consider the influence of longitudinal wind on the hydrogen leakage trajectory. In this paper, the computational fluid dynamics (CFD) software STAR CCM 2021.1 is used to simulate the hydrogen leakage and diffusion trajectories of fuel cell vehicles (FCVs) at five different leakage locations the longitudinal wind speeds of 0 km/h, 37.18 km/h and 114 km/h, and it is concluded that longitudinal wind prolongs the diffusion time of hydrogen to the headspace and reduces the coverage area of hydrogen in the headspace with a decrease of 81.35%. In order to achieve a good detection effect of fuel cell vehicles within the longitudinal wind scene, based on the simulated hydrogen concentration–time matrix, the scene clustering method based on vector similarity evaluation was used to reduce the leakage scene set by 33%. Then, the layout position of HCSs was optimized according to the proposed multi-scene full coverage response time minimization model, and the response time was reduced from 5 s to 1 s.

Detection of optic disk (OD) in fundus image is a standard procedure before the diagnosis and analysis of various serious ophthalmic pathologies. Manual detection will consume both time and energy. Therefore automatically detection is rather necessary and practical. However, owing to uneven illumination, disorder distribution of optic vessels or noises in the fundus images, this task has constantly been challenging. In this paper, a novel method based on Gaussian vessel detector with contrast tuning, morphology operations and tangent information, is presented to detect OD from retinal fundus images. After converting the color image to grayscale image by its green channel, image contrast is tuned intended for the subsequent vessel detection. Proceeded with Gaussian vessel detector, the search space for the later operation will be restricted to a limited dimension. Next, morphology operations as well as denoising operations guarantee that interferences of trivial vessels or noise eliminated. Eventually, by applying the tangent information transformation on the pre-processed image, the center of the Optic disc is determined. This method achieved a success ratio of 91.3 (74 out of 81 images) in the standard database S.T.A.R.E. The experimental result shows that the proposed method is effective and could approximate the average accuracy of a human observer.