Numerical Method (China)

This paper presents a digital-twin-based model predictive control framework for autonomous process control, demonstrated in a virtual experiment on thermal food processing in a convection oven. In combination with prior work, this approach enables simulation-centered food scientists to deploy physics-based simulation models in live process control environments. The digital twin is realized as a physics-based, data-driven reduced-order model (ROM) that provides faster-than-real-time predictions. The ROM is trained on trajectories from a high-fidelity multiphysics finite-element model of chicken fillets. A central contribution is a model predictive control scheme that overcomes the common fixed-initial-condition limitation of augmented neural ordinary differential equation ROMs: a dedicated sub-optimization step re-synchronizes the surrogate to the measured state of the food item at each control instant, allowing reliable live re-optimization without access to internal ROM states. The controller optimizes oven temperature setpoints to meet target food-quality metrics (core temperature, moisture content, texture) and autonomously accommodates changes to the planned end time during operation. Quantitatively, the ROM achieves relative time-series errors of 0.18–0.49%, and the control algorithm evaluates 501 trajectories of 1800 s real time in a total of 46.6 s on a single core of a processor, demonstrating on-device feasibility without cloud or edge resources. Receding-horizon model predictive control of the remaining setpoints mitigates model–reality mismatch, enforces user-defined food metrics, and sustains closed-loop performance under autonomous operation. • Software-agnostic pipeline to deploy digital-twin surrogates for online control. • Accurate, faster-than-real-time predictions enable autonomous operation. • Online sub-optimization re-synchronizes predictions to measured core temperature. • Level 5 autonomy case beyond the trained time window. • On-device digital twin operation possible without cloud or edge resources.

In semi‑supervised multi‑view classification (SMVC), scarce labels and noisy unlabeled data impair feature aggregation and compromise prediction reliability, while existing methods lack principled guidance and interpretability. To overcome these limitations, we propose a novel unified SMVC framework, Neural Collapse Priors Driven Trust Semi-Supervised Multi-View Classification (NCPD-TSMVC), building upon neural collapse–derived prototype priors and evidential opinion fusion. Concretely, we rigorously prove under neural collapse theory that normalized classifier weights from the labeled‑data pre‑training stage coincide with class centroids in feature space, conferring maximal inter‑class separation and optimal within‑class compactness. These prototype priors permeate the entire learning pipeline, calibrating the representation learning of unlabeled samples to obtain highly discriminative embeddings. Simultaneously, our evidential learning module quantifies epistemic uncertainty and fuses view‑level opinions at the evidence level, yielding robust and transparent decision making. Extensive evaluations across diverse benchmarks demonstrate that NCPD‑TSMVC surpasses state‑of‑the‑art SMVC approaches in performance, robustness and interpretability.

In dynamic and complex real-world environments, artificial intelligence (AI) vision systems continue to face significant challenges in accurately detecting and tracking small objects. The core difficulty lies in the fact that small targets usually exhibit limited spatial and textural features, while dynamic backgrounds often generate numerous misleading motion cues, thereby interfering with reliable discrimination between targets and backgrounds. Inspired by the remarkable capability of the insect brain in detecting small moving objects, this study proposes a visual neural network model enhanced by a feedback mechanism. By adaptively responding to temporal variations, the proposed model is able to more precisely distinguish small targets from background-induced false targets. The network architecture consists of two main pathways: a motion detection pathway that extracts motion-related features from minute targets, and a feedback attention pathway that enhances the focus on true targets by leveraging the feature differences between real and false motion signals. In addition, a global inhibition module is incorporated to reduce the false alarm rate by filtering out background-induced false positives, thereby improving the overall detection performance of the model. Experimental results demonstrate that the proposed model achieves a detection rate of 0.81 in complex visual scenarios, whereas the compared models all achieve detection rates below 0.59, indicating a significant improvement in detection performance. Meanwhile, in terms of Precision and F1-score, the proposed model achieves values of 0.0648 and 0.12, respectively, while the compared models obtain values lower than 0.0077 and 0.015, further validating the superiority of the proposed method in detection accuracy and robustness.

In this paper, we propose a two-stage Deep Reinforcement Learning (DRL) framework for the automatic arrangement of buildings in urban planning, with a focus on satisfying critical regulatory constraints: sunshine duration and inter-building distance. To overcome the computational bottleneck of high-fidelity sunshine simulation during DRL training, we first introduce a Convolutional Auto-Encoder (CAE) model that serves as a fast and accurate surrogate for predicting sunshine distribution across a building layout. This CAE model, trained on data generated from Autodesk Revit, achieves near real-time prediction (∼0.015 s per layout) with an average error of less than 3 min, enabling its seamless integration into the iterative RL loop. Subsequently, we formulate the sequential placement of buildings as a Markov Decision Process (MDP) and employ a Deep Q-Network (DQN) to learn the optimal placement policy. The DQN agent interacts with a planning environment, where its actions (moving a building in discrete directions) are rewarded based on the satisfaction of distance constraints and the avoidance of low-sunshine zones identified by the CAE. Our integrated CAE-DQN system automates the generation of compliant building layouts. Experimental evaluation demonstrates that our method significantly outperforms baseline approaches (Random Placement, Greedy Heuristic, and Genetic Algorithm) in terms of success rate (100%), convergence speed, and layout compactness. The proposed framework not only ensures hard constraint satisfaction but also provides a practical and scalable tool for rapid prototype generation in the conceptual design phase of urban planning.

The environments of severe thunderstorms that produced hail were examined using 171 proximity soundings (2005-2024) archived in the 3rd Data Release of Project SWAP. These soundings were categorized based on their geographical occurrence into three hail-prone environments across Luzon, Philippines. For each case, key parameters describing instability, vertical wind shear, and moisture were calculated to assess the environmental conditions for hail production. The probability of hail occurrence, expressed as a function of WMAX (√2 × CAPE) and 0-6 km bulk shear (DLS), revealed patterns distinct from those reported in other regions. Hail events in Luzon were most likely under high CAPE conditions, where boundary-layer moisture was sufficient, mid- and low-level lapse rates were steep, and lifting condensation levels were high. Surprisingly, weak DLS was common across Luzon hail environments, diverging from existing severe weather climatologies, yet large DCAPE indicated environments conducive to damaging wind events. When DLS was replaced with the shear magnitude between the cloud base and equilibrium level, the probability of hail occurrence increased, better aligning with global severe weather climatologies. This finding is supported by hodograph analyses, which show largely unidirectional wind profiles: strong speed shear aloft but weak directional shear in the low-levels. Parameters such as WMAXSHEAR, WMAXSHEARLCL-EL, and BWDLCL-EL emerge as potential discriminators between non-severe and severe thunderstorms capable of producing hail, and as useful metrics for assessing convective storm severity in Luzon and possibly countrywide. Finally, two recurring severe setups conducive to hail were identified: (1) an easterly regime associated with trade winds, and (2) a westerly regime linked to the Asian summer monsoon.

Reconstructing physical fields from sensor measurements is a challenging task in scientific and engineering applications, especially when sensors are sparse and unevenly distributed. Existing generative models directly incorporate point observations into the generation process to guide the generation of the full field. However, they did not provide equitable guidance for relatively sparsely observed regions, resulting in an imbalanced reconstruction error. To mitigate this imbalance, we propose proper orthogonal decomposition conditioned flow matching (POD-FM), a framework that synergistically integrates gappy POD's global physical consistency propagation to provide reconstruction guidance for poorly observed regions. Specifically, we take the POD modal coefficients inverted from observations as physically interpretable global feature guidance, then jointly encode them with the point observations to guide the generative model to reconstruct the full field. These coefficients provide a low-dimensional global prior that anchors the probability flow to the dominant modes of the field, which supplements implicit physical consistency for sparsely observed areas, and thus alleviates the accuracy imbalance caused by uneven sensor distribution. The efficacy of POD-FM has been verified on three progressive POD modal complexity datasets. Experiments have shown that POD-FM achieved higher accuracy and could effectively suppress the influence of sensor noise. The code will be available at https://github.com/XinWang199/POD-FM.

Cross-modal hashing (CMH) aims to bridge the semantic gap between heterogeneous modalities by learning compact binary representations for efficient retrieval. Most existing deep cross-modal hashing methods are developed under the assumption that multimodal data are complete and perfectly paired across modalities. However, this assumption rarely holds as real-world multimodal datasets often suffer from missing modalities due to inconsistencies, imbalances, or noise during data collection. To address such incomplete data, existing incomplete CMH methods typically attempt to reconstruct the missing information by exploiting internal signals from the available modalities. Nonetheless, these internally guided completion strategies tend to be highly sensitive to distributional shifts, leading to substantial performance degradation on unseen or out-of-distribution data. Inspired by the human learning mechanism of enhancing cognition through external knowledge, this paper proposes a novel External Guidance Incomplete Cross-modal Hashing (EGICH) framework to address this limitation. Specifically, we first design a Completion with External Guidance (CEG) module that leverages rich semantic information from external knowledge bases to expand the semantic boundary and accurately reconstruct the semantics of missing samples. Subsequently, we introduce a Consistency Learning with External Guidance (CLEG) module, which employs externally guided reconstructed features as anchors to align sample representations with label semantics, thereby effectively mitigating cross-modal bias. Finally, a Semantic-aware Contrastive Hashing (SCH) module is developed to refine the feature distribution by semantic similarity, pulling semantically related samples closer and pushing unrelated ones apart, thus achieving fine-grained discrimination among positive pairs. To the best of our knowledge, this is the first attempt to incorporate external knowledge into incomplete cross-modal hashing. Extensive experiments demonstrate that EGICH consistently and significantly outperforms 11 state-of-the-art methods under various modality-missing scenarios. The code is available at https://github.com/chenjiali27/EGICH.

A. q-Rung orthopair fuzzy sets (qROFS) constitute a robust extension of Intuitionistic Fuzzy Sets (IFS) and Pythagorean Fuzzy Sets (PyFS).They provide an advanced framework for handling complex information in an uncertain environment.The REGIME method is an MCDM approach used to provide a definitive ranking of alternatives.In the literature, a few extensions have integrated REGIME with IFS and PyFS.However, not only do very few of these extensions address group decision-making problems, but they are also limited when dealing with q-ROFS type data.In this article, we propose an extension of the REGIME method to group decision-making in a qROFS environment.Our methodology proceeds in three phases: aggregation of fuzzy decision matrices, formulation of a non-linear programming model based on qROFS entropy for the objective weighting of criteria, and adaptation of REGIME to evaluate the options.This new methodology extends, for the first time, the REGIME method to a qROFS environment.The proposed method is applied to a numerical example on supplier selection to demonstrate its effectiveness.