This work proposes a hybrid computational framework that integrates the Virtual Element Method (VEM) with neural network techniques to efficiently solve three-dimensional hyperelastic problems. The framework combines graph neural networks (GNNs), VEM, and physics-informed neural networks (PINNs), and incorporates a transfer learning strategy to improve the model generalization ability. Owing to its ability to operate on arbitrary polyhedral meshes, VEM provides a flexible and accurate discretization for hyperelastic problems involving complex geometries. Building on this property, the proposed framework employs VEM to construct high-fidelity physics constraints, utilizes GNNs to encode the topology of polyhedral meshes, and adopts PINNs to solve the governing equations under these constraints. Furthermore, the introduced transfer learning strategy enables rapid adaptation of the trained model to new material parameters and loading conditions. Numerical experiments demonstrate that the proposed framework achieves accuracy comparable to that of traditional high-fidelity numerical simulations while improving computational efficiency by one to two orders of magnitude. In addition, the method exhibits strong robustness and generalization capability under variations in parameters and loading scenarios. These results indicate that the proposed approach provides a promising and efficient solution for the rapid analysis of multi-scenario and multi-parameter three-dimensional hyperelastic problems.
Vehicle re-identification (Re-ID) plays a vital role in vehicle-infrastructure cooperation systems, enabling continuous trajectory association across non-overlapping roadside camera views and providing essential perception redundancy for autonomous driving. However, in complex urban environments, frequent occlusions caused by surrounding vehicles, roadside structures, and pedestrians severely degrade Re-ID performance. Existing external model–based solutions often depend on auxiliary segmentation or detection networks, which increase computational overhead and hinder deployment in resource-constrained roadside units. In contrast, model-free approaches based on image enhancement provide limited gains when addressing structural occlusions, while multi-image feature fusion methods require additional images for joint training, substantially increasing training complexity. To address these challenges, this study proposes a novel dual-branch feature reconstruction model for the Re-ID (DFRR) framework, which integrates a dual-branch image feature encoder for discriminative feature extraction and an image reconstruction module for occluded image restoration under diverse and enhanced occlusion scenarios. Without relying on external models or neighboring image information, DFRR effectively recovers structural details in occluded regions, enabling robust single-image Re-ID in real-world traffic conditions. Extensive experiments on the VeRi-776 and DAIR-V2XReid datasets demonstrate the effectiveness of the proposed framework, achieving a higher mean average precision and delivering competitive performance compared with benchmark methods, particularly under severe occlusion conditions.
Time-dependent partial differential equations (PDEs) underpin modeling of dynamical phenomena across science and engineering, but repeatedly solving them at high fidelity remains expensive. Neural operators offer reusable surrogates for such systems, yet current approaches often lose robustness under sparse or irregular temporal supervision and can be costly to train for high-resolution or long-trajectory problems. Here we introduce the Time-Attentional Neural Operator (TANO), a composite neural operator whose stacked layers each extract multiscale spatial features, model their temporal interactions with attention, and reconstruct the spatial representation. We further develop a stochastic temporal subsampling strategy in which each training step uses only a random subset of time points, improving training efficiency while retaining one-pass full-trajectory prediction over the supervised horizon. Across five canonical PDE benchmarks, TANO reduces prediction errors by 1.87–3.94× relative to existing baselines and remains robust under sparse temporal supervision or limited data. It accurately simulates wave propagation over long trajectories in heterogeneous media, trains about 4× faster than the competing approach on high-resolution problems, and can be extended to long-horizon extrapolation through a windowed rollout scheme. Our findings establish TANO as an efficient and robust neural-operator framework that makes training on complex spatiotemporal problems more computationally tractable and supports practical surrogate modeling.
Atmospheric CO2 emissions are driving the climate crisis, prompting research into catalytic hydrogenation to produce value-added chemicals. Here, a ternary CoCuAl mixed-metal oxide catalyst derived from collapsed layered double hydroxides (LDHs) was developed for selective CO2 hydrogenation to C5+ hydrocarbons at 240 degrees C. Compared with LDH-derived CuAl (mainly yielded CO and methanol) and CoAl (mainly methane) catalysts, the ternary CoCuAl catalyst with a Co/(Co + Cu) ratio of 0.53 exhibited a remarkable shift in selectivity toward C5+ hydrocarbons (12.2%) due to synergy between Cu and Co. Cu promoted CO2-to-CO conversion via the reverse water-gas shift (RWGS) reaction, while Co enabled subsequent chain growth through Fischer-Tropsch synthesis (FTS). The low temperature of 240 degrees C was essential, as higher temperatures likely cause Cu-Co segregation and metallic Co particles that favor methanation over FTS. Importantly, the LDH-derived catalyst outperformed the conventional co-precipitated CoCuAl analog, delivering higher CO2 conversion (16.3% vs. 9.5%) and C5+ selectivity (12.2% vs. 5.0%), owing to superior metal dispersion from the LDH confinement effect. XPS confirmed strong Co-Cu interactions, and in-situ IR spectroscopy revealed key CO and formate intermediates. This work highlights the potential of LDH-derived mixed oxides for efficient CO2 valorization under mild conditions.
This study examines how CCR schemes targeting ride-sourcing services influence multi-modal travel behavior in systems with public transit, ride-sourcing, and private vehicles to promote green transportation and reduce congestion and emissions. By designing Credit-Charge-cum-Reward scheme targeting ride-sourcing services, the study seeks to balance government fiscal revenue, intervention objectives, and ride-sourcing platform profitability. To rigorously capture the policy-constrained interaction between the ride-sourcing platform and travelers, a bi-level Stackelberg game-theoretic framework is developed. Within this framework, the government establishes the CCR policy environment as an exogenous intervenor. Under these policy constraints, the ride-sourcing platform acts as the leader in the upper-level problem by optimizing pricing and driver supply strategies, while travelers act as followers in the lower-level problem by making equilibrium mode choices to minimize generalized travel costs. The results demonstrate that CCR schemes effectively shift travelers toward green modes and reduce system-wide emissions while maintaining overall traffic efficiency. Interventions targeting public transit and private cars can incentivize a modal shift from high-emission private driving to ride-sourcing services. Furthermore, the study reveals the complementary functions of different credit mechanisms under fiscal neutrality, where interventions on private vehicles and transit directly determine emission reduction while those on ride-sourcing ensure fiscal neutrality in the regular equilibrium regimes, and incentive-based strategies further enhance platform profitability. Consequently, the proposed mechanism effectively reconciles the conflicts among government fiscal balance, platform viability, and traffic efficiency improvements.
