Search Results (372)

We introduce and study a class of multi-composite sigmoid neural network operators for Banach space-valued approximation. The proposed operators are generated by density-type kernels induced by finite compositions of seven standard sigmoid-type activation functions. The approximation is considered for continuous functions on compact intervals of the real line and on the whole real line, with values in an arbitrary Banach space $(X,\parallel ·\parallel )$. We prove quantitative pointwise and uniform convergence results by means of Jackson-type inequalities expressed through the first modulus of continuity. Higher-order and fractional approximation results are also obtained in terms of Banach space-valued derivatives and Caputo–Bochner fractional derivatives. The associated feed-forward neural network representation has one hidden layer and uses the multi-composite sigmoid function as its activation. Numerical experiments are presented to validate the theoretical estimates and to illustrate the approximation behavior of the proposed operators. In particular, we compare classical tanh-based operators, normalized self-composed activation operators, and heterogeneous multi-composite activation operators. The results show that self-composition and heterogeneous composition may improve the uniform approximation error for certain activation families and parameter choices, while also indicating that the observed improvement is activation-dependent and influenced by the composition order, kernel localization, and the regularity of the target function. Full article

High dynamic range (HDR) imaging offers an enhanced visual experience by capturing a wider range of real-world luminance levels in digital images. Driven by the increasing demand for high-quality visuals, HDR monitor technology has seen significant advancements. As such monitors become commonplace in both consumer and professional settings, efficient methods are needed for both converting standard dynamic range (SDR) content to HDR—known as reverse tone mapping—and optimizing natural HDR lighting content for display on HDR monitors. A reverse tone mapping procedure aims to produce natural lighting levels, but even on high-end HDR monitors, such images still require adjustment to avoid hard clipping. This paper presents a solution that jointly does both steps: (1) reverse tone mapping to a display-aware HDR representation, and (2) direct generation of an image tailored for a chosen monitor brightness value. We propose a novel neural network architecture conditioned on the target peak brightness via a lightweight multi-layer perceptron (MLP) module injected at the bottleneck, which predicts a bracketed stack of LDR exposures serving as the method’s HDR representation. In this manner, the ill-posed tone mapping problem is guided by auxiliary information about display characteristics, improving visual quality. Experiments throughout the full consumer HDR range (100–4000 nits) show consistent improvements over the display-agnostic baseline in peak luminance utilization, local contrast, color and perceptual quality. Full article

This paper presents a novel, efficient computational framework for the optimisation of the fundamental frequency of multi-layered composite slabs with consideration of uncertainties. The approach is based on Finite Element Method (FEM) data generation, Deep Neural Network (DNN) surrogate modelling, deterministic optimisation using the genetic algorithm (GA), Morris Sensitivity Analysis (SA), and quantile-based optimisation, including uncertainties and using the GA. Different boundary condition configurations are considered. The surrogate model is trained on FEM-generated samples and subsequently used to replace expensive modal analyses during optimisation, significantly reducing the optimisation evaluation cost for one boundary condition variant. The proposed method achieves near-identical optimal non-dimensional parameter $\mathsf{\Omega }$ values to those reported in the literature for Bayesian Optimisation (BO), with discrepancies of less than 0.5%. To improve robustness to manufacturing tolerances, an additional uncertainty-aware optimisation is performed, in which model parameters are perturbed with normally distributed noise. By maximising the 5% quantile of the non-dimensional parameter $\mathsf{\Omega }$, robust optimal solutions are obtained with minimal loss in performance. Overall, the DNN-GA framework enables fast and accurate optimisation of composite laminates and provides both deterministic and robust design recommendations at a fraction of the computational cost of traditional FEM-based optimisation workflows. Full article

This work is devoted to the application of complex-valued neural networks based on the multilayer neural network with multi-valued neurons (MLMVN) for short-term electrical load forecasting in smart grid energy systems. Accurate forecasting is a critical component of energy management systems, as it directly impacts the efficiency of control and optimization strategies in increasingly distributed and stochastic environments. The proposed approach leverages the intrinsic properties of complex numbers to model periodicity and nonlinear relationships typical of load time series. A compact feedforward architecture with two hidden layers is adopted and combined with multiple preprocessing strategies, including unit circle encoding, Fourier transform representations, and hybrid feature mappings incorporating temporal information such as the day of the week. The performance of the proposed models is evaluated on real-world prosumer data and compared against two benchmarks: a seasonal persistence model and a Long Short-Term Memory network. Results show that MLMVN-based approaches achieve comparable or improved performance in terms of RMSE and error reduction capability, despite their lower architectural complexity. Fourier-based preprocessing methods demonstrate strong effectiveness in capturing underlying temporal patterns. These findings suggest that complex-valued representations provide a promising alternative to traditional deep learning approaches, offering a favorable balance between accuracy, interpretability, and computational efficiency in Smart Grid forecasting applications. Full article

Pig weighing is an essential procedure for monitoring growth and animal health; however, conventional methods are often labor-intensive, costly, and potentially stressful. In this context, this study proposes a non-invasive approach for estimating the body weight of pigs during the growing stage based on computer vision and the YOLOv11 algorithm, enabling automatic segmentation and individual identification in multi-animal environments. The study used RGB images of 10 group-housed pigs captured throughout the growing phase, in which automatic dorsal segmentation was combined with individual identification through numerical markings. From the generated binary masks, the segmented dorsal area was extracted and used as a predictor variable in Linear Regression and a Multilayer Perceptron (MLP) Artificial Neural Network. The YOLOv11 model showed consistent performance in the segmentation task, achieving test-set metrics of Precision = 0.849, Recall = 0.886, mAP@0.50 = 0.936, and mAP@0.50–0.95 = 0.819, demonstrating good generalization capability in scenarios with intense animal interaction. In the weight prediction stage, Linear Regression and the MLP achieved high coefficients of determination (R² = 0.96 and 0.95, respectively) with low errors (RMSE = 1.52 kg and 1.63 kg; MAE = 1.20 kg and 1.25 kg), indicating a strong correlation between segmented dorsal area and actual body weight. Class-wise analysis revealed superior performance for classes 7 and 9, with R² values up to 0.98 and RMSE below 1.1 kg, whereas class 8 showed greater error dispersion, associated with higher morphological variability and a smaller number of available samples. These results demonstrate that the direct use of morphometric information extracted from segmented masks in 2D images constitutes a robust, accurate, and low-cost approach for automatic pig body-weight estimation. Moreover, this study is among the few addressing this task specifically during the growing stage, highlighting its potential for future deployment in embedded systems and intelligent monitoring platforms for precision pig farming, although further evaluation of computational efficiency and real-time performance is still required. Full article

In air jet looms, relay nozzles are critical components in governing airflow velocity and air consumption during the weft insertion process. Although computational fluid dynamics (CFD) offers high-fidelity simulation for aerodynamic analysis, its computational burden hinders its practicality in iterative aerodynamic design of relay nozzles. To address the challenge, this study proposes a data-driven framework integrating a Chebyshev polynomial Kolmogorov–Arnold Network (Chebyshev KAN) surrogate model with an Improved Multi-objective Red-billed Blue Magpie Optimizer (IMORBMO). The accuracy of the Chebyshev KAN model was benchmarked against conventional multilayer perceptrons (MLP), convolutional neural networks (CNN), and the standard Kolmogorov–Arnold Network (KAN). Experimental results demonstrate that the Chebyshev KAN model achieves the lowest mean absolute error (MAE) of 0.103 for airflow velocity and 0.115 for air consumption. Building upon the non-dominated sorting and crowding distance strategies, IMORBMO was developed, incorporating an adaptive mutation mechanism by information entropy for improvement of convergence, diversity, and uniformity of the Pareto-optimal solutions. Comprehensive evaluations on the ZDT and WFG benchmark suites confirm that the IMORBMO consistently attains the best and highly competitive performance, yielding the lowest generation distance (GD), inverted generational distance (IGD) values and the highest hypervolume (HV). Applied to the aerodynamic optimization of a relay nozzle, the proposed framework delivers an optimal aerodynamic design that increases airflow velocity by 10.5% while reducing air consumption by 15.4%, as verified by CFD simulation. The steady-state flow field was simulated by solving the Reynolds-Average NavierStokes equations with the k–ω turbulent model, utilizing Fluent 2025.R2. No-slip wall, inlet pressure and outlet pressures are boundary conditions to the relay nozzle surfaces. This work establishes a computationally efficient and accurate optimization paradigm that holds significant promise for aerodynamic design and other complex real-world engineering applications. Full article

The dynamic response of the secondary circuit system in nuclear propulsion plants is critical to the power output, safety, and energy efficiency of nuclear-powered ships. High-fidelity thermo-hydraulic simulation models can accurately capture system transients but are computationally expensive and unsuitable for real-time applications. To address this limitation, this study proposes a reduced-order dynamic parameter prediction method that integrates high-fidelity simulation with deep learning. A multi-operating-condition simulation model of a typical nuclear-powered ship secondary circuit system is developed to generate time-series data covering load ramping and propulsion mode switching. Based on this dataset, a conventional recurrent neural network (RNN) and a multilayer long short-term memory (LSTM) network are constructed for multivariate autoregressive prediction of 17 key dynamic parameters, and their performances are systematically compared. Results show that the LSTM significantly outperforms the RNN in capturing long-term temporal dependencies, achieving average RMSE and MAPE values of 0.0228% and 0.365%, respectively. The proposed model completes 50-step-ahead prediction within 0.84 s, satisfying real-time requirements. The hybrid simulation-driven and data-driven framework provides a practical solution for intelligent monitoring and control optimization of nuclear-powered ship propulsion systems. Full article

Deep learning models have achieved high accuracy in forecasting problems, but at the cost of large computational energy demand. Brain-inspired spiking neural networks (SNNs) offer a promising, low-power alternative, yet their adoption for time-series forecasting has been limited by information loss from binary spikes and degraded performance in deeper networks. This paper proposes a fully spiking framework that bridges this gap by improving both the encoding and propagation of information in SNNs. The framework introduces a hybrid Delta-Rate encoding mechanism that captures both abrupt changes and gradual trends in time-series data, and a Mem-Spike mechanism that transmits analog membrane potential values to preserve fine-grained information between spiking layers. We further employ residual membrane connections to maintain signal flow in deep spiking networks. Using two public energy load datasets, our enhanced SNNs consistently outperform conventional spiking models, improving prediction accuracy by up to 61.6% and mitigating degradation in multi-layer networks. Notably, it narrows the gap to the selected deep learning baseline (LSTM), achieving comparable accuracy in some settings while requiring only about 10% of the estimated inference energy of that baseline under a common operation-level model. These results show that, within the empirical scope considered here, enhanced conventional SNNs can improve time-series forecasting accuracy while retaining favorable estimated efficiency. Full article

Hybrid electric mining trucks operating in open-pit environments encounter highly variable gradients and payload conditions that standard energy management strategies fail to address adequately. Existing approaches are predominantly calibrated for full-load scenarios and neglect the accelerated battery degradation resulting from sustained high-power cycling, undermining long-term operational viability. This study presents a multi-objective energy management framework that couples real-time driving condition recognition with dynamic programming (DP) optimization for a 130-tonne hybrid mining truck. Field data collected from an open-pit mine in Heilongjiang Province were used to construct six physically representative driving conditions via principal component analysis and K-means clustering. A Bidirectional Gated Recurrent Unit (Bi-GRU) network (2 layers, 128 hidden units per direction) was trained on a route-based temporal split, attaining 95.8% classification accuracy across all six conditions. Condition-specific powertrain modes were subsequently defined, and a DP formulation with a weighted-sum cost function was solved to jointly minimize diesel consumption and battery capacity fade—quantified through a semi-empirical effective electric quantity metric. A marginal rate of substitution (MRS) analysis was conducted to identify the optimal trade-off between fuel economy and battery life preservation. In the DP cost function, the weight coefficient μ (ranging from 0 to 1) governs the relative emphasis placed on battery degradation minimization versus fuel consumption minimization: μ = 0 corresponds to pure fuel minimization, whereas μ = 1 corresponds to pure battery degradation minimization. The MRS analysis identified μ = 0.1 as the knee point of the Pareto trade-off: relative to pure fuel minimization (μ = 0), this setting reduces effective electric quantity by 6.1% while increasing fuel consumption by only 1.4% (MRS = 4.36). Against a rule-based baseline, the proposed strategy improves fuel economy by 12.3% and extends battery service life by 15.7%. Co-simulation results were validated against onboard fuel-flow measurements; absolute simulated and measured fuel consumption values are reported route-by-route, with deviations within 4.5%. A three-layer BP neural network (3 inputs, two hidden layers of 20 and 10 neurons, 1 output) trained on the DP solution reproduces near-optimal performance—with fuel consumption and effective electric quantity increases below 1.0% and 1.1%, respectively—while reducing computation time by over 96% (from approximately 52,860 s to 1836 s for the 1800 s driving cycle), demonstrating practical feasibility for real-time deployment. Full article

Aiming to address the problem of extracting the remote sensing FTIR spectral characteristics of the hot jet of a certain type of aero-engine under different working conditions, this paper proposes a feature construction algorithm for the remote sensing FTIR spectral characteristics of the aero-engine hot jet based on the fusion of the original spectral features and the deep spectral features. The infrared spectrum was collected at a distance of 280 m, covering the spectral range of 2.5–15 μm with a resolution of 1 cm⁻¹. The Neighborhood–Autoencoder Integration Dual-Branch Network (NAIDN) feature construction algorithm is proposed. This algorithm contains a neighborhood integration branch and an autoencoder branch. The neighborhood integration branch converts the radiation intensity values of discrete wavenumber points into local energy aggregation features through a sliding window, accurately extracting the key physical information in the original spectrum. The autoencoder branch uses a three-layer fully connected neural network architecture to mine the deep spectral features of the spectral data. The algorithms of the two branches not only retain the physical interpretability of spectral analysis but also capture the multi-parameter coupling information hidden in the hot jet spectrum through the representation learning ability of the autoencoder, achieving feature fusion across spatial dimensions. Compared with traditional feature construction algorithms, the dual-branch feature construction algorithm proposed in this paper has stronger comprehensive representation capabilities. The content of carbon dioxide (CO₂) and cyanide groups (-C≡N) in the hot jet under different operating conditions varies significantly. In the experiment, an unsupervised clustering algorithm, the Agglomerative Clustering classifier, is selected, and the classification accuracy of the features extracted by the algorithm in this paper reaches 92.97% on this classifier, thereby verifying the effectiveness of the algorithm in this paper. Full article

Modeling the dynamic response of saturated marine soils is crucial yet computationally challenging for traditional methods. Meanwhile, purely data-driven models suffer from sparse data and lack of physical interpretability. To overcome these limitations, this study proposes an intelligent engineering framework based on a frequency-domain physics-informed neural network (FD-PINN) for the forward simulation and inverse parameter identification of saturated seabed soils. Constrained directly by physical laws during the learning process, FD-PINN remains highly reliable even when training data is sparse. By formulating the governing equations in the frequency domain, it directly predicts complex-valued displacement and pore-pressure phasors. Multiscale Fourier feature mappings mitigate spectral bias and capture boundary layers and high-frequency effects. For inverse problems, a phase-sensitive lock-in extraction strategy transforms time-domain measurements into robust frequency-domain targets, enabling the accurate and noise-tolerant identification of poroelastic parameters with clear physical meaning (nondimensional storage parameter S and permeability parameter $\Gamma$). Numerical experiments show that FD-PINN substantially outperforms conventional time-domain PINN, achieving relative $L^2$ errors of ${10}^{-2}\sim {10}^{-3}$ for single- and multi-frequency excitations typical of wave-induced loadings. In particular, $\Gamma$ is consistently recovered with sub-percent relative error, while S can be reliably identified with multi-frequency data. The framework offers a data-efficient, noise-robust approach for high-fidelity modeling and robust parameter inversion, which is particularly valuable in offshore environments where high-quality data is scarce. Full article

Fiber dyeability is a core indicator of textile quality and added value. Pre-experiment accurate prediction of fiber dyeability reduces the waste and inefficiency of trial-and-error methods. However, due to the limited data volume and complex mechanisms of fiber dyeability, there are no relevant studies to date. Thus, this paper proposes a novel prediction model integrating domain knowledge and process data called multi-head attention–physics-informed neural network (MHA-PINN). Within the MHA-PINN framework, limited experimental data is first augmented by using variational autoencoders, and subjected to ensemble feature selection on the augmented samples. Subsequently, a multi-head attention layer is introduced to capture the interdependencies among sample variables, thereby outputting a new feature matrix that represents the weighted fusion of these variables. Finally, a physics-informed neural network module embeds the dyeing kinetic equations into the loss function, guiding the model to converge towards accurate solutions for sample predictions. The effectiveness and superiority of the proposed MHA-PINN have been validated on a fiber dyeability experimental dataset. Full article

The unique ecological gradients of Xinjiang have fostered a rich reservoir of genetic resources in local sheep populations. However, the population genetic structure, adaptive mechanisms to extreme environments, and the genetic basis underlying key economic traits of these breeds remain poorly understood. To address this gap, we performed whole-genome resequencing of 140 individuals from seven indigenous sheep populations—Altay, Bayinbuluke, Kazakh, Kirgiz, Bashibai, Turpan Black, and Yemule White—identifying 18,700,507 high-quality SNPs. Genetic diversity analyses revealed that all populations exhibited comparable levels of genetic diversity, with modest variation across breeds, with Turpan Black sheep exhibiting the highest observed heterozygosity (Ho = 0.3110) and proportion of polymorphic sites, whereas Kirgiz sheep showed comparatively lower values. Population structure analyses consistently indicated that geographic isolation is the primary driver of genetic differentiation, with Kirgiz sheep from the Pamir Plateau in southern Xinjiang displaying the greatest genetic distance relative to northern Xinjiang populations. By integrating multiple selection signature detection methods—including F_ST, π ratio, and XP-CLR—we found that genes under selection in Kirgiz sheep were significantly enriched in biological pathways related to stem cell pluripotency regulation (e.g., BMPR1B), DNA repair (e.g., DDB2), and neural development, thereby elucidating their unique genetic adaptations to high-altitude environments. In contrast, Turpan Black sheep appear to cope with heat stress through mechanisms involving basal transcriptional regulation (e.g., GTF2I), maintenance of protein homeostasis (e.g., DNAJB14), and melanin biosynthesis (e.g., MC1R). Furthermore, comparative analysis of body size identified a suite of candidate genes associated with growth and development (e.g., CUX1, KIT), which are primarily involved in transcriptional regulation, protein kinase activity, and the ubiquitin-mediated proteolytic system, thereby revealing a multi-layered genetic regulatory network governing body conformation. Collectively, this study provides a comprehensive genomic framework for understanding the genetic structure, adaptive evolution, and molecular basis of economically important traits in indigenous sheep breeds from Xinjiang, offering valuable candidate targets for future functional validation and precision breeding programs. Full article

The deep integration of machine learning technology with geological prospecting has brought to the forefront a key challenge: how to construct geological-mineralization models by fusing multi-source data, select model features with guidance from metallogenic factors, build multi-source metallogenic prediction models with geological constraints, and ultimately achieve a thorough integration of domain knowledge and machine intelligence. The Eastern-Central Tianshan region is one of China’s most important copper–nickel mineral resource bases, predominantly hosting magmatic copper–nickel sulfide deposits with significant resource potential. In this context, this paper proposes a metallogenic prediction model based on multi-modal feature fusion technology. The model employs a Residual Neural Network (ResNet) incorporating a Squeeze-and-Excitation (SE) attention mechanism and a Multi-Layer Perceptron (MLP) to extract features from different modalities. It integrates multi-source data, including geochemical information, geological metallogenic factors, and aeromagnetic data. A cross-modal feature interaction module, constructed using attention weighting and a gating mechanism, enables deep fusion of the features. After training, the model achieved a prediction accuracy of 97% on the test set. Compared to a unimodal model constructed using Random Forest, the confidence and discriminative capability of the training results were significantly enhanced, validating the effectiveness of multi-modal feature fusion. Applying the trained model to the study area, a total of 11 prospective metallogenic zones were delineated. These include 4 zones in the peripheries of known deposits and 7 zones in previously unexplored (blank) areas. Notably, some known mineral occurrences fall within the predicted blank-area targets, validating the feasibility and significant value of multi-modal feature fusion in mineral prediction. This work provides a novel methodology for the subsequent integrated processing of multi-source data. Full article

Background/Objective: Breast density and cancer risk are key imaging-derived biomarkers, yet their assessment is limited by inter-reader variability and inconsistent reproducibility. This Technical Note evaluates the feasibility of a bifurcated neural network designed to simultaneously predict breast density and a composite cancer risk index, providing a methodological foundation for future integration into contrast-enhanced mammography (CEM) workflows. Materials and Methods: A simulated cohort of 1000 patients was generated to reproduce clinically plausible variability in breast density (Densitanum) and cancer risk (RiskEnum). A multi-output neural network was developed and compared with two baselines: multiple linear regression and a single-output multilayer perceptron (MLP). Performance was assessed using R², mean squared error (MSE), and mean absolute error (MAE). Learned trends were examined for consistency with established physiological and epidemiologic patterns. Results: Linear regression showed limited explanatory power (R² ≈ 0.144). The single-output MLP improved prediction of the cancer risk index (R² = 0.436; MSE = 9.558). The bifurcated neural network achieved MAE values below 4 for both outputs (2.624 for Densitanum; 3.731 for RiskEnum), demonstrating robust performance and the advantage of simultaneous multi-target prediction. The model reproduced clinically coherent patterns, including the expected age-related decline in breast density. Conclusions: This simulation-based feasibility study demonstrates that bifurcated neural networks can jointly model correlated breast imaging biomarkers with high internal consistency. The proposed architecture provides a reproducible methodological platform that can be directly tested on real CEM datasets to support future AI-enhanced risk stratification and personalized screening pathways. Full article