搜索结果：convergence

For non-Pieper robotic manipulators that lack closed-form analytical solutions, inverse kinematics (IK) is typically solved using numerical or optimization-based methods. However, conventional Particle Swarm Optimization (PSO) algorithms often suffer from premature convergence, unstable performance near singular configurations, and limited trajectory smoothness. To address these challenges, this paper proposes an improved PSO framework that integrates a nonlinear inertia weight strategy with Jacobian pseudo-inverse guidance. The nonlinear inertia weight employs a two-phase adaptation mechanism, enhancing global exploration in the early iterations and improving local exploitation later, thus avoiding stagnation. The Jacobian guidance introduces gradient-based directional information, which accelerates convergence and improves robustness under singular conditions. Furthermore, a quaternion-based pose error formulation combined with a joint continuity penalty ensures smooth trajectory tracking and avoids orientation singularities. Experimental results on benchmark functions and two representative non-Pieper manipulators demonstrate that the proposed method achieves superior convergence accuracy, improved stability, and better trajectory continuity compared with multiple PSO variants. The results confirm the effectiveness and generality of the proposed framework for solving IK problems of non-Pieper robots.

This paper proposes an integral sliding mode based adaptive robust backstepping control scheme to improve the trajectory tracking and hovering performance of a quadrotor unmanned aerial vehicle (UAV) under large-scale time-varying disturbances. It also considers the impact of variations in the payload mass of the UAV, such as in tasks such as power line maintenance or rescue operations. The proposed scheme effectively mitigates the influence of disturbances on the flight process when the upper bound of the time-varying disturbance is unknown, and estimates the potentially uncertain parameters of the system in real time. Using Lyapunov stability theory, it was proven that the designed controller ensured the asymptotic convergence of the tracking error to zero. Furthermore, this paper integrates adaptive control with the concept of integral sliding mode, combining their respective technical characteristics in a complementary manner. The proposed adaptive law, incorporating a σ-modification term, effectively suppresses the chattering inherent in sliding mode control, ensuring system stability. The integration of the sliding mode surface further accelerates the error convergence to zero. The simulation results validate the performance of the proposed control scheme in various scenarios, including continuous weak disturbances, changes in payload mass, and sudden large-scale time-varying disturbances. The results demonstrate that the proposed control scheme has strong robust stabilization and wide applicability, outperforming the traditional adaptive robust control methods and classical PID methods.

Early diagnosis of plant leaf diseases plays an important role in protecting crop yields and supporting sustainable agriculture. This paper proposes an improved DeepFusionNet model optimized through a hybrid Flower Pollination Algorithm and Butterfly Optimization Algorithm, balancing global exploration with local refinement for faster and more stable convergence. The model combines DenseNet201 and MobileNetV2 by compressing their final convolutional feature maps with 1×1 convolutions and fusing them along the channel dimension to form a compact and discriminative representation. This fused representation is then classified using a Random Forest classifier. This framework consistently achieves high accuracy on all eight datasets, with performance ranging between 97.07% and 99.66%. Extensive experiments are performed that include statistical validation, convergence studies, and reliability tests to prove the robustness of the approach. Furthermore, to make it practically useful, the whole system is embedded into a mobile application capable of real-time disease detection and providing actionable recommendations to farmers for the effective treatment and prevention of diseases.

This paper addresses the challenging problem of adaptive prescribed-time tracking control with guaranteed performance for uncertain strict-feedback nonlinear systems subject to arbitrary initial errors. Unlike existing methods that require restrictive initial conditions or are limited to stabilization problems, this work develops a unified control framework that simultaneously achieves three key objectives: handling arbitrary initial errors without performance violation, ensuring exact prescribed-time convergence, and maintaining guaranteed transient performance throughout the operation. The main contributions include a novel generalized tanh-based error transformation technique that decouples control performance from initial error magnitudes, automatically confining arbitrary initial errors within asymmetric prescribed performance bounds. Furthermore, an enhanced adaptive backstepping framework incorporating fractional-order terms is proposed to guarantee that the tracking error enters the prescribed performance region before a user-specified entry time and converges to zero exactly at a prescribed settling time. Rigorous theoretical analysis demonstrates the uniform ultimate boundedness of all closed-loop signals while achieving both prescribed-time convergence and guaranteed transient performance under system uncertainties. Simulation results confirm the effectiveness of the proposed method.

The increasing penetration of photovoltaic distributed generation (PV-DG) in Radial Distribution Systems (RDSs) plays a vital role in achieving sustainable energy transition objectives; however, the inherent uncertainty associated with solar irradiance and load demand poses significant challenges to optimal planning and operation. This paper presents a stochastic optimization framework for PV-DG allocation in RDSs using the Barrel Theory-Based Optimizer (BTO). Uncertainties in solar irradiance and load demand are explicitly modeled using appropriate probability density functions and efficiently represented through a higher-order Point Estimate Method (PEM), which captures the essential statistical characteristics with a limited number of representative scenarios. The proposed framework simultaneously optimizes the location and capacity of PV-DG units to minimize real power losses and enhance voltage profile performance while ensuring system operational constraints are satisfied. The effectiveness of the proposed approach is validated on the 85-bus and the IEEE 118-bus RDSs, where the BTO exhibits superior convergence characteristics and enhanced solution robustness when compared with several benchmark optimization techniques, including the well-established Differential Evolution Algorithm (DEA), the recent Crocodile Ambush Optimization (CAO, 2025), and the Schrödinger Optimizer Algorithm (SOA, 2025). For the 85-bus RDS, the impact of integrating different numbers of PV units is systematically investigated. Simulation results confirm that the proposed BTO-based stochastic planning strategy significantly improves energy efficiency, voltage regulation, and loss reduction, thereby enhancing the overall sustainability of the RDS. For the 85-node RDS, the BTO achieves a noticeable reduction in average real power losses, outperforming DEA, CAO, and SOA by 2.55%, 4.10%, and 6.74%, respectively, when three PV units are installed. Additionally, for the case of four PV units, the proposed BTO yields even greater improvements, with loss reductions of 5.12%, 7.50%, and 14.12%, respectively, compared with the same benchmark algorithms. Furthermore, for five PV units, the BTO achieves much greater reduction, outperforming DEA, CAO, and SOA by 13.05%, 6.45%, and 32.31%, respectively, when three PV units are installed.

Representatives of the phylum Methanobacteriota occur in various anoxic environments, but only members of the genera Methanosphaera and Methanobrevibacter exclusively colonize the digestive tract of animals. Recent phylogenomic analyses revealed that the genus Methanobrevibacter, which harbors the majority of the intestinal species, is severely underclassified and represents a family-level taxon, "Methanobrevibacteraceae", that evolved entirely in the digestive tract of animals. Comparative genome analysis of 158 species of Methanobacteriota, including uncultured representatives in the Genome Taxonomy Database (GTDB), demonstrated that the intestinal lineages are clearly separated from the remaining members of the phylum. They differ from the non-intestinal lineages in genome size, GC content, coding density, an increased number of pseudogenes and adhesin-like proteins, and show numerous adaptations to the copiotrophic gut environment. A decreased biosynthetic potential led to a dependence on other community members and limits the dispersal of intestinal species into other habitats, which is reflected in coevolutionary patterns with their major host groups among arthropods, ungulates, and primates. Certain lineages even engaged in symbiotic associations with intestinal protists, presumably benefiting from the H2 produced by the hydrogenosomes of their anaerobic hosts. Our results reveal that the transition of free-living Methanobacteriota to a host-associated lifestyle involves the same genomic changes that were previously recognized in gut bacteria and bacterial endosymbionts of protists, reflecting resemblances between the two prokaryotic domains that are caused by evolutionary convergence in similar environments.

Clinical decision support systems (CDSS) increasingly rely on artificial intelligence (AI) to interpret biomedical images and provide accurate, real-time diagnostics. Federated Learning (FL) has emerged as a promising privacy-preserving solution for training AI models across decentralized healthcare institutions. However, FL performance is significantly affected by data heterogeneity-especially when client data is non-independent and identically distributed (non-IID), a common occurrence in clinical practice. In this study, we propose a dynamic FL framework specifically designed for Retinal Vein Occlusion (RVO) detection that enhances the adaptability and robustness of CDSS. The core contribution of our approach is a server-side aggregation strategy that selects FedAvg for IID data and SCAFFOLD for non-IID data based on monitoring the normalized L2 divergence between successive communication rounds. This intelligent selection mechanism ensures stable convergence, improves diagnostic accuracy, and supports reliable model performance across diverse healthcare environments. Experimental results show that the proposed approach was evaluated on a two-class RVO dataset across five simulated clients under both IID and non-IID data distributions. By addressing real-world data distribution challenges, this work offers a practical path toward integrating FL-based AI systems into clinical workflows for more effective and trustworthy medical image analysis.

Embryonic development relies on precise temporal regulation of signaling pathways, yet most computational analyses of protein-protein interactions treat signaling events as static and time-invariant. Here, we present a time-resolved, ensemble-based docking framework to examine stage-dependent patterns in relative interaction propensity within the canonical Wnt/β-catenin signaling pathway. Core pathway interactions spanning extracellular, membrane-associated, cytoplasmic, and regulatory components were analyzed using conformational ensembles generated without molecular dynamics simulations. To integrate structural variability with developmental context, we defined a Stage-Weighted Interaction Score (SWIS), which combines ensemble docking-derived structural compatibility, pose convergence, and stage-specific relevance weighting into a normalized comparative metric. SWIS is designed for relative evaluation of interaction features across stages and does not represent binding affinity, interaction stability, or kinetic properties. Application of this framework revealed stage-dependent differences in relative interaction propensity across early, mid, and late signaling phases. Receptor-proximal interactions exhibited higher composite scores during early stages, whereas intracellular adaptor and effector interactions showed increased relative prominence during intermediate stages. Late-stage analysis indicated reduced composite scores across the network. Network-level analysis further revealed stage-dependent redistribution of interaction features under context-specific weighting, rather than persistence of a fixed interaction architecture. These patterns reflect context-weighted redistribution of docking-derived interaction features rather than direct changes in intrinsic biochemical interaction stability. Together, this framework provides a scalable computational approach for comparative analysis of temporally contextualized signaling interactions and generates testable hypotheses regarding the temporal organization of developmental signaling networks.

Emotional facial expressions are known to bias face processing, yet it remains unclear whether such effects extend to neural signals associated with face familiarity. In this secondary analysis of openly available EEG data, we used cross‑dataset multivariate pattern analysis (MVPA) to test whether established neural signatures of face familiarity generalize across emotional expressions. Participants viewed faces in two independent experiments: one involving explicit emotion categorization (happy, angry, sad, neutral) and another involving personally familiar and unfamiliar identities. Classifiers trained to distinguish familiar from unfamiliar faces were cross‑applied to emotional expressions, and vice versa, using complementary relabeling strategies. Across analyses, neural patterns for angry expressions showed the strongest and most sustained generalization to familiarity‑related neural signals, emerging around 200 ms post‑stimulus and persisting throughout the trial (peak Cohen's d = 1.35 over posterior regions). Neural patterns for happy and sad expressions showed weaker and more transient generalization (200-400 ms), while neutral expressions consistently aligned with patterns for unfamiliarity. These findings demonstrate that threat‑related facial expressions exhibit neural dynamics that show convergence in pattern structure with established familiarity signals, extending prior evidence that emotional expressions, particularly anger, systematically modulate face representations beyond identity.

Traditional aptamer screening methods often prove ineffective for small molecule targets, primarily due to the inherent structural limitations of such compounds. Their simple architecture, limited functional groups, and restricted spatial complexity drastically reduce the probability of identifying nucleic acid sequences that bind with both high affinity and specificity. Consequently, the screening process becomes inefficient and labor-intensive, frequently failing to yield aptamers of satisfactory performance for practical applications. This represents a significant technical hurdle in expanding the use of aptamers in small-molecule detection and therapeutics. Based on this, this study innovatively proposes an aptamer design method based on single-nucleotide docking assembly, using the small molecule temicloxacin as an example. Through molecular dynamics simulations (50 ns, RMSD convergence threshold of 0.15 nm), the dynamic conformational characteristics of tilmicosin were analyzed. Subsequently, saturated docking was performed on four classes of mononucleotides, screening out 32 high-affinity mononucleotides (atomic contact distance ≤4 Å). Methods such as depth-first search algorithm (DFS) and weighted graph theory model were introduced to obtain the representative single nucleotides of eight classes of functional modules and linkage assembly, and finally 63 non-redundant candidate sequences were screened. Molecular docking results indicate that the optimal aptamer Til-14 exhibits high binding affinity with tilmicosin. with an affinity of 298.16 ± 95.588 nM measured via SYBR Green I fluorescence assay. Colloidal gold colorimetric analysis confirmed its high affinity (Kd = 279.323 ± 87.234 nM) and excellent specificity. This innovative method successfully addresses the key limitations of the traditional SELEX process in screening aptamers for small molecule targets. By enhancing the efficiency and specificity of selection, it not only facilitates the discovery of high-performance aptamers but also establishes a novel, generalizable framework for the construction of nucleic acid aptamers targeting other small molecules.

This study aimed to systematically review functional neuroimaging literature on the neural substrates underlying contextual modulation of pain in healthy individuals. A search was conducted in PubMed-Medline, Cochrane, and Web-of-Science databases (PROSPERO-CRD42024586392). Studies on chronic pain were excluded, and the risk of bias was assessed with Cochrane RoB2. Spatial coordinates of brain regions undergoing activity changes were included in a meta-analysis using both Activation Likelihood Estimation and frequency estimation of activated or deactivated regions in individual studies (convergence analysis). From 224 full texts reviewed and n=100 articles retained (2735 individuals), three broad activity patterns were identified. One involved activation of prefrontal cortical areas, together with a modification of the sensory message in nociceptive cortical areas (deactivated during hypoalgesia and hyperactivated in hyperalgesia), consistent with top-down regulation via descending controls. A second configuration also involved prefrontal mobilisation, but without activity changes in nociceptive areas and was consistent with a 'perceptual decision bias'. A third configuration was associated with irregular ventromedial prefrontal involvement and significant deactivation in dorsolateral and ventrolateral prefrontal areas. While the first two patterns were observed in a range of attentional or expectation manipulations, including placebo/nocebo, the third pattern was essentially observed during tasks involving introspection and self-referential procedures such as meditation, religious prayer or nostalgia. Similar subjective pain changes can coexist with different brain activation patterns, reflecting diverse neural strategies. Whereas prefrontal cortex-driven descending modulation is one mechanism, introspective approaches can alter perception without involving prefrontal activity. This mechanistic diversity supports multiple avenues for behavioural or neuromodulation-based pain control.

The proliferation of hate speech on social media poses a significant challenge to maintaining safe and inclusive online environments, necessitating accurate and scalable automated detection systems. However, the performance of transformer-based models for hate speech detection is highly sensitive to hyperparameter configurations, making manual and conventional tuning strategies inefficient in high-dimensional search spaces.To address this challenge, this study proposes a hybrid optimization framework that integrates the DeBERTaV3 transformer model with the Grey Wolf Optimizer (GWO) for automated hyperparameter tuning. The proposed approach enables efficient exploration of complex hyperparameter spaces by balancing global search and local refinement. The framework optimizes eight critical hyperparameters, including learning rate, weight decay, and dropout rates, to enhance convergence stability and generalization performance. The proposed method is evaluated on the Davidson et al. (2017) dataset, consisting of 24,783 labeled tweets. Experimental results demonstrate that the GWO-DeBERTaV3 model achieves a peak accuracy of 97.72% and a macro F1-score of 97.71%, with statistically significant improvements over baseline and conventional tuning approaches.These findings highlight the effectiveness of metaheuristic-based optimization for transformer fine-tuning and demonstrate its potential for improving robustness and performance in real-world hate speech detection systems.

Federated learning (FL) has become a highly promising paradigm for privacy-preserving distributed model training by enabling edge devices to train without sharing raw data. But in practice, edge environments are both non-stationary and asymmetric, with varying data distributions due to shifts in user behaviour, sensing conditions, and overall environmental dynamics. This causes concept drift (sudden, gradual, and recurrent), leading to poor model performance, slower convergence, and predictive bias. Current approaches to FL are not combined to tackle problems of drift adaptation, differential privacy (DP) and resource efficiency (FedAvg, DP-FedAvg). To address these constraints, we present FedDriftGuard. This Federated learning layer unifies client-level drift detection, drift-adaptive aggregation, and adaptable differential privacy into a single, FLE architecture-compatible system. The proposed DP-DriftNet model implements attention-based time encoding to capture changing data patterns and drift-directed feature weighting to allow greater flexibility in the presence of distributional changes. A drift-optimal privacy scheduler allocates noise probabilistically, subject to a limited privacy budget, thereby enforcing an appropriate privacy-utility trade-off without cancelling formal DP guarantees. Also, update sparsification, compression and periodic transmission techniques are used to reduce communication overhead. Decades of experimentation on real-world and synthetic drift datasets have shown that FedDriftGuard outperforms baseline FL techniques, achieving accuracy and F1-score gains of 9-14% and 11-17%, respectively, with adaptation latency 28% shorter and communication cost 20-35% lower. Such findings are statistically significant and confirm the soundness of the suggested method. FedDriftGuard offers effective, scalable privacy-preserving learning in adaptable, edge-drifting environments.

Ischemic stroke poses a substantial clinical and socioeconomic burden due to limited therapeutic efficacy and poor neurological outcomes. To uncover novel gene targets for intervention, we conducted an integrative analysis combining single-cell RNA sequencing with Mendelian randomization using large-scale genomic datasets from the European Bioinformatics Institute (34,593 cases and 624,214 controls), with validation in an independent European Bioinformatics Institute dataset (86,668 cases and 1,503,898 controls) and the UK Biobank (26,052 cases and 487,214 controls). Colocalization analysis identified four core genes-PEBP1, BMP4, APOA1 and CD86-strongly associated with ischemic stroke risk, with a posterior probability of a shared causal variant greater than 0.8. Among them, PEBP1 was markedly upregulated post-ischemia, particularly in endothelial cells, as confirmed by quantitative PCR and immunofluorescence in a middle cerebral artery occlusion model. Both pharmacological inhibition of PEBP1 with FerroLOXIN-1 and AAV-BI30-mediated shRNA knockdown reduced cerebral infarct volume, enhanced neuronal survival, and improved neurological functional recovery. In vitro, FerroLOXIN-1 enhanced cell proliferation and viability under oxygen-glucose deprivation conditions, with potential off-target effects of the interventions validated. Mechanistically, these effects were mediated through activation of the Akt/p38 MAPK signaling cascade. These findings highlight PEBP1 as a central mediator of ischemia-induced neuronal injury and a potential therapeutic target. The convergence of transcriptomic, genetic and experimental validation supports the translational relevance of PEBP1 inhibition in post-stroke neuroregeneration.

Chagas disease (ChD) and Type 2 diabetes (T2D) originate from distinct etiological processes -infectious and metabolic, respectively- yet both share a chronic inflammatory and metabolic imbalance that profoundly impacts immune-endocrine homeostasis. Persistent Trypanosoma cruzi infection in ChD induces sustained immune activation, altered adrenal steroid balance, and tissue remodeling, whereas T2D is characterized by metabolic inflammation, oxidative stress, and insulin resistance. When these two conditions coexist, their overlapping inflammatory, metabolic, and endocrine circuits may act synergistically, amplifying metabolic toxicity, immune exhaustion, and premature immunosenescence. In addition, this comorbidity thus represents the convergence of pathogen-driven and metabolism-driven inflammation, resulting in a disrupted neuroendocrine-immune dialogue and heightened susceptibility to tissue damage, particularly in the heart. Understanding the mechanistic basis of this interplay is crucial, as it highlights shared pathogenic pathways and potential molecular targets for integrated therapeutic interventions. Altogether, recognizing ChD+T2D coexistence as a mechanistic rather than merely epidemiological association provides new insights into the links between chronic infection, metabolic dysfunction, and immune aging-offering a conceptual framework for future studies aimed at restoring immune-metabolic balance and improving disease outcomes, particularly cardiac damage.

Accurate detection and segmentation of moving objects constitute a fundamental challenge in computer vision, particularly for intelligent video surveillance systems operating under variable illumination, dynamic backgrounds, and environmental noise. This paper presents a fully unsupervised dual-phase motion analysis framework that effectively combines statistical independence modeling and geometric contour evolution to achieve high-precision motion detection and segmentation. In the first phase, an enhanced Fast Independent Component Analysis (Fast-ICA) algorithm is employed to perform statistical decomposition of video sequences, exploiting temporal independence to distinguish moving foregrounds from static backgrounds. This process generates an initial motion mask with strong robustness to illumination variation and noise artifacts. In the second phase, a hybrid level set segmentation model integrating the global Chan-Vese formulation and a locally adaptive Yezzi-based energy function refines object boundaries through an adaptive energy minimization process. A stabilization term and a self-regulating convergence criterion are further incorporated to ensure contour smoothness, numerical stability, and resilience to topological changes. Comprehensive experiments conducted on the CDNet-2014 benchmark dataset demonstrate that the proposed method achieves an average recall of 0.9613, precision of 0.9089, and F-measure of 0.9310, outperforming several state-of-the-art supervised, semi-supervised and unsupervised background subtraction algorithms. The proposed Fast-ICA-Level Set fusion framework thus provides a robust, adaptive, and computationally efficient solution for real-world intelligent surveillance and autonomous visual monitoring applications.

In this manuscript, we investigate the problem of prescribed-time bearing-based formation tracking for multi-agent systems. The proposed control law employs a two-stage strategy to achieve formation tracking within a prescribed time. For multi-agent systems with time-varying leader velocities, follower agents estimate the leaders' inputs through a prescribed-time bearing-based observer. The second stage of the control law is a prescribed-time bearing-based formation tracking controller. Relying solely on the measurement and communication of bearing-related information between neighboring agents, the controller drives the followers to achieve the desired bearing-constrained formation in a prescribed time for first-order systems and enables the agents to achieve velocity coordination with the leaders for second-order systems. The prescribed times for the two stages of the control law can be assigned independently by the user, and convergence is proved using Lyapunov analysis. In addition, to demonstrate the effectiveness and practical feasibility of the proposed control law, simulations and a UAV swarm experiment are conducted.

In view of the challenges such as feature redundancy and insufficient small sample risk prediction accuracy caused by the proliferation of multi-source heterogeneous data in the distribution network in the ubiquitous power Internet of Things environment, this study proposes a distribution network risk prediction method based on data mining and improved swarm intelligence algorithm fusion support vector machine, namely the DM-IS model. First, this study uses kernel principal component analysis technology to map high-dimensional nonlinear data to low-dimensional space to eliminate redundant feature. Then, the nonlinear adaptive weights and chaotic mutation strategy of the improved particle swarm optimization algorithm based on logistic mapping are introduced. A hybrid model that can accurately optimize the penalty factors and kernel parameters of support vector machines globally is constructed. The results revealed that this model not only effectively overcome the premature convergence defect in the parameter optimization process, but also achieved high-precision fitting with a median error of only 0.022 in 92.0 s under extreme small-sample constraints, outperforming baseline models by significantly narrowing the error distribution bandwidth. In actual engineering scenarios, its average detection time for typical faults such as transformer overload was shortened by 77.90%, and monthly operation and maintenance costs were reduced by 42.65%. In addition to confirming the efficacy of multi-source heterogeneous data fusion in enhancing forecast robustness, this study offers quantitative algorithm reference and decision support for conversing power systems from passive repair to active defense by continuously quantifying risk probability indices and dynamic early-warning time margins to drive predictive parameter adjustments and early equipment replacements prior to critical systemic failures.