Scipedia: Revista Internacional de Métodos Numéricos para Cálculo y Diseño en Ingeniería

Hierarchical Convolutional Neural Network for Emotion Recognition Using EEG and Facial Expressions

Scipedia content — Mon, 04 May 2026 11:18:13 +0200

Emotion recognition is crucial for advancing human–computer interaction (HCI) by enabling systems to interpret complex affective states. While Electroencephalogram (EEG) signals provide direct insights into neural activity, facial expressions offer external emotional cues. However, unimodal systems often struggle with robustness and generalization across diverse subjects. This study presents a Hierarchical Convolutional Neural Network (HCNN) framework that integrates EEG and facial expressions through multi-level convolutional feature extraction and featurelevel fusion. The proposed model combines deep hierarchical representations with handcrafted temporal–frequency and texture-based descriptors to form a unified feature vector. Experiments on the MAHNOB-HCI and DEAP datasets show that the HCNN achieves accuracies of 91.40% and 88.09%, outperforming CNN-, LSTM-, and SVM-based methods. The results demonstrate the model’s ability to effectively capture complementary cross-modal correlations while reducing feature redundancy and computational complexity. The HCNN framework shows great promise for real-time emotion recognition applications, offering a scalable, interpretable, and data-efficient solution for multimodal emotion recognition in next-generation HCI systems.

Hybrid Fractional Coupled Systems with Generalized Caputo Derivatives

Scipedia content — Fri, 08 May 2026 09:55:14 +0200

This paper establishes a comprehensive theoretical framework for a novel coupled system of nonlinear hybrid fractional differential equations involving generalized Caputo derivatives. The system’s hybrid nature, coupled with the generality of the fractional operators, allows for modeling complex interdependent processes with memory effects that cannot be adequately captured by existing models.
Using Krasnoselskii’s fixed point theorem, we prove the existence of at least one solution under explicit coupling conditions. Under appropriate Lipschitz conditions, we establish uniqueness via Banach’s contraction principle, deriving a quantitative condition involving the fractional orders, generalization parameters, and Lipschitz constants. We also conduct a rigorous analysis of Ulam-Hyers and Ulam-Hyers-Rassias stability, obtaining explicit stability constants that provide quantitative bounds on how perturbations in the input affect the solution.
The theoretical results are validated through three carefully constructed numerical examples with explicit parameter values demonstrating existence, uniqueness, and Ulam-Hyers stability. A parameter sensitivity analysis confirms the robustness of the uniqueness condition across variations in fractional orders, generalization parameters, and interval lengths. The paper concludes with a discussion of limitations and directions for future research, including extensions to higher fractional orders, delay and impulsive effects, and numerical methods.

Efficient Mean Estimation Using Repeated Measurements in PPS Sampling with Scientific Applications

Scipedia content — Mon, 08 Jun 2026 10:12:03 +0200

Research studies in many scientific disciplines need efficient estimation methods for estimating the parameters of quantitative variables, including, but not limited to, tensile strength, wind speeds, air quality, and temperature, etc. Statisticians employ a sampling design to get a random sample and estimate the population means based on the observed sample. If the units of a finite population have different selection probabilities, unequal probability methods of sample selection are used. The existing unequal probability sampling methods allocate probabilities proportional to size (PPS) to the population units by using a one-time measurement approach. Memory– type estimation methods, on the other hand, use multiple measurements and provide more precise estimates than the traditional estimators. This research study introduces a novel memory-type estimator using a PPS sampling design. Various properties of the suggested mean estimator are analyzed, and the efficiency conditions are derived. A few real-world data sets have been used from previous studies to evaluate the performance of the suggested and competing mean estimators. Our comparative study suggests that the suggested memory-based estimator performs much better than the competing estimators, which means that the suggested memory-type estimator is an appropriate estimator for application in real-life sample surveys.

Enhancing Few-Shot Text Classification with Parameter-Efficient Tuning of Large Language Models

Scipedia content — Mon, 08 Jun 2026 10:13:03 +0200

Traditional few-shot text classification models focus only on label prediction and cannot extract structured information such as entities or events, limiting their usefulness in real-world, semantics-driven tasks. They also rarely use external knowledge or parameter-efficient tuning, leading to shallow representations and weaker performance. To address this, this paper proposes a knowledge-aware multi-task framework that integrates few-shot classification with entity and event extraction. A single BERT encoder with IA3 adapters enables efficient tuning, while semantic triples extracted via spaCy and aligned with WordNet and ConceptNet are encoded using TransE. A BiLSTM captures sequential context and a softmax decoder performs token-level prediction. Experiments show strong results—97.97% accuracy, 98.00% precision, 97.95% recall, and 97.96% F1—surpassing state-of-the-art baselines. Ablation studies confirm the value of the knowledge-enhanced, multi-step design, demonstrating suitability for lowresource, knowledge-centric applications.

AI-IoMT Synergy: A Real-Time Framework for Automated Urinary Tract Infections (UTI) Detection Based on Urine Sediments

Scipedia content — Mon, 08 Jun 2026 10:15:03 +0200

A Urinary Tract Infection (UTI) is characterized by an infection affecting the urinary system, including the kidneys, bladder, urethra, and ureters, with clinical presentations including pyelonephritis, cystitis, and urethritis. While conventional diagnostic methods such as urinalysis and urine culture and sensitivity (C&S) remain widely used, they are limited by subjectivity, time-intensive processing, susceptibility to contamination and risks of false-positive or false-negative results. This study proposes a comprehensive deep learning (DL) and Internet of Things (IoMT) framework to automate the real-time detection and classification of UTIs using microscopic urine sediment images. The study employed 2 datasets (A and B). Dataset A, a clinically acquired dataset, comprises of 3345 images (normal, erythrocytes, fungi and pus) and Dataset B, a publicly accessible data comprises of 5377 images and 26,419 cropped microscopic images corresponding to 7 classes (casts, crystals, erythrocytes leukocytes, epithelial cells, epithelial nuclei, mycetes). A two-stage classification approach was implemented: a binary task to distinguish urine sediments from normal cases, followed by a multiclass task (clinical data and online data) to identify the specific infection type. All images underwent pre-processing, including normalization, resizing, noise removal, and augmentation to enhance feature visibility and model generalizability. The data were partitioned into training (65%), validation (25%), and test (10%) sets. Six state-of-the-art DL architectures, including ResNet50-V2, ResNet101-V2, Inception-V3, XceptionV2, Inception-ResNet-V2, and VGG19 were fine-tuned using transfer learning and evaluated using accuracy, precision, recall, F1-score, and confusion matrices. The proposed models were uploaded to a website to enable realtime detection (accessible via this link: https://uticlassification.app/). The proposed pipeline demonstrated strong performance in both classification tasks, underscoring the potential of deep learning as a reliable, rapid, and reproducible tool for automated urine sediment identification in clinical practice.

Carbon-Aware Last-Mile Delivery Optimization using Sparrow search algorithm and graph neural network risk assessment

Scipedia content — Thu, 11 Jun 2026 09:11:14 +0200

With the advancement of global carbon neutrality strategies and explosive growth in e-commerce, urban last-mile delivery faces multiple challenges in balancing economic benefits, environmental impact, and risk management. Traditional optimization methods struggle to simultaneously address multi-objective trade-offs, network topological dependencies, and small-sample risk prediction problems. This study proposes a hybrid intelligent framework integrating an Improved Sparrow Search Algorithm (ISSA) with a Meta-Learning Graph Convolutional Network on Prototype Space (ML-GCNPS) for carbon-aware delivery optimization and risk assessment. ISSA-NSGA-III generates high-quality initial populations through Tent chaotic mapping, designs an adaptive periodic convergence factor to dynamically balance exploration and exploitation, and enhances global search capability by integrating Lévy flight with Elite Opposition-Based Learning (EOBL), achieving multi-objective collaborative optimization through embedding in the NSGA-III framework. ML-GCNPS designs a feature extraction network to extract discriminative features from multi-modal node data, explicitly models class centers through prototype space embedding to enhance small-sample generalization, constructs an adaptive Vertex-to-Edge (V2E) network to dynamically infer edge weights and capture risk propagation paths, and employs a two-layer graph convolutional architecture for sufficient information propagation. Experiments on the Kaggle Supply Chain Management dataset and Carbon Monitor risk dataset demonstrate that compared to standard SSA, ISSA-NSGA-III improves total cost, carbon emissions, and resource utilization by 14.0%, 14.2%, and 15.4%, respectively, with Pareto front quality improved by 28.5%. ML-GCNPS achieves an AUPRC of 0.850 (standard deviation 0.005) and a Macro Fl-Score of 0.850 (standard deviation 0.005) in 5-way 1-shot scenarios, reduces the False Negative Rate (FNR) to 0.080 (standard deviation 0.003), and achieves a Weighted Average Cost (WAC) of 45, significantly outperforming baseline methods such as ProtoNG and MAML-GNN (paired t-test, p less than 0.01 for AUPRC and FNR versus the second-best baseline GAT-FSL). Ablation experiments validate the necessity of the meta-learning framework, graph convolutional structure, V2E network, and prototype space embedding, while alternative design comparisons demonstrate the superiority of the technical choices. While the individual algorithmic components (Tent chaotic mapping, Lévy flight, EOBL, prototypical networks, and GCN) are established techniques, the principal contribution of this study líes in their systematic integration into a closed-loop optimization-assessment-feedback decision framework, where graph structure serves as an information bridge connecting delivery optimization with risk prediction. This study provides a teclmical solution for smart city logistics and offers theoretical basis and practica! guidance for sustainable delivery under carbon neutrality goals. All reported improvements over baseline methods are statistically significant across all evaluation metrics (paired t-test or Wilcoxon rank-sum test, p < 0.001

Rainfall Trend Analysis and One-Dimensional Hydrodynamic Assessment of the Ganga River Basin Using HEC-RAS 6.3

Scipedia content — Thu, 09 Jul 2026 11:54:54 +0200

This study integrates long-term rainfall trend analysis with one-dimensional hydrodynamic modelling to assess hydro-climatic variability and flood vulnerability along the selected stretch of the Ganga River near Farakka Barrage, eastern India. Daily gridded rainfall data for 1991–2025 were analyzed to identify trends and potential change points. The Mann–Kendall test, Sen’s slope estimator, Bias-Corrected Pre-Whitening, and Trend-Free Pre-Whitening were applied to account for serial correlation. The rainfall series showed an overall decreasing tendency, although statistical significance varied among the applied trend tests after pre-whitening. HEC-RAS 6.3 simulations showed increasing water surface elevation, velocity, and discharge under 25-, 50-, and 100-year design-flood scenarios. The findings indicate that flood risk in the study reach is shaped by both hydro-climatic variability and local hydraulic controls.

Multivalued Fixed Point Methods for Sequential Coupled Caputo-Type Fractional Differential Inclusions with Coupled Boundary Conditions

Scipedia content — Mon, 20 Jul 2026 13:29:03 +0200

We study a coupled system of sequential Caputo-type fractional differential inclusions with two-point integral boundary conditions. Using fixed-point techniques for multivalued operators, existence results are established under Carathéodory-type and Lipschitz-type assumptions on the multifunctions. The obtained results extend several known solvability criteria for fractional differential inclusions with nonlocal boundary conditions. An example is included to illustrate the applicability of the theoretical results.

Dynamic Mechanisms and Multiparameter Control of Fracture-Induced Lost Circulation in Deep Wells—A Fluid-Solid Coupling Approach

Scipedia content — Tue, 25 Aug 2026 12:26:04 +0200

Deep and ultra-deep well drilling in multi-scale fractured formations frequently encounters fracture-induced lost circulation, which restricts drilling efficiency and reservoir development. To address this problem, this paper establishes a dynamic finite element numerical model based on fluid-solid coupling theory, revealing the multi-physics coupling mechanism of fracture propagation and leakage channel evolution. The model fully considers the seepage-stress coupling effect during drilling fluid circulation, and quantitatively analyzes the influences of pump rate, fluid density, fluid viscosity, fracture geometric parameters, elastic modulus, Poisson’s ratio and in-situ stress on fracture aperture and fluid loss rate. The research results show that: pump rate and fluid density are positively correlated with loss rate and fracture aperture; fluid viscosity presents two-stage regulation characteristics, which is negatively correlated with leakage parameters at the fracture initiation stage and positively correlated in steady-state leakage stage; longer fracture length increases loss rate but reduces fracture aperture; elastic modulus, Poisson’s ratio and in-situ stress mainly affect the early leakage behavior, with little influence on steady-state stage. The proposed multi-parameter coupling control model can realize dynamic prediction of multi-fracture leakage in deep wells, and provides theoretical support for on-site anti-leakage and plugging work.

A Robust Ensemble Learning Paradigm for Breast Cancer Diagnosis Based on Multi-Classifier Performance Optimization

Scipedia content — Tue, 25 Aug 2026 12:22:07 +0200

Breast cancer is among the major causes of cancer-related mortality in women, and early diagnosis is critical to improved survival. However, traditional diagnostic methods or standalone models often lack precision and have the potential to give false-negative results. This study aims to present a hyperparameter-tuned ensemble learning model for the classification of breast benign and malignant tumors based on the Wisconsin Breast Cancer (Original) dataset. Data quality was ensured through a series of preprocessing techniques: median imputation, Zscore standardization, and the Local Outlier Factor (LOF) noise reduction algorithm. The classical Machine Learning (ML) models (SVM, Random Forest (RF), XGBoost, Bagging) were compared with the Deep Learning models (VGG16 and DenseNet121). A hybridized soft-voting ensemble of the four classical classifiers (SVM, RF, XGBoost, and Bagging) was then constructed, and its hyperparameters were tuned with the OPTUNA framework using the Treestructured Parzen Estimator (TPE), a Bayesian optimization algorithm, over 30 trials. The results of the experiments demonstrated that the Ensemble + OPTUNA model performed with an accuracy of 99.25% and a recall of 1.00 for the malignant class compared to the individual models and recent state-of-the-art methods. This is an optimized ensemble for reliable and computationally stable performance. To enhance transparency and privacy-preserving validation in clinical settings, Explainable AI (XAI) will be incorporated into future research, while Federated Learning will be enhanced to better validate the model.

A Numerical Modeling Framework for Hydraulic Performance Evaluation and Sensitivity Analysis of Underground Pump Station Schemes for Multistage Centrifugal Pumps

Scipedia content — Mon, 14 Sep 2026 12:09:23 +0200

With the increasing application of pumping systems in energy and industrial fields, accurate hydraulic performance evaluation and parameter analysis have become essential for improving the design and operation reliability of pump stations. This study develops a CFD-based numerical modeling framework for the hydraulic assessment and sensitivity analysis of underground pump station schemes equipped with multistage centrifugal pumps. The Reynolds-averaged Navier–Stokes (RANS) approach combined with the RNG k-ε turbulence model was employed to establish three-dimensional numerical models of different centrifugal pump configurations. Based on the obtained numerical results, the hydraulic performance characteristics, internal flow behaviors, and engineering applicability of three pump schemes were systematically evaluated. Furthermore, a modified Morris screening method integrated with the Latin hypercube sampling–partial rank correlation coefficient (LHS–PRCC) method was incorporated into the numerical framework to quantify the influence of six hydraulic parameters on pump head and efficiency. The results demonstrate that the single-suction vertical three-stage volute centrifugal pump with conventional flow channels exhibits the best overall hydraulic adaptability for the investigated underground pump station. The impeller outlet width shows the highest sensitivity to pump head, whereas flow rate is the dominant factor affecting efficiency. The diffusion section height of the volute has a relatively limited influence on both head and efficiency. The proposed numerical framework provides a systematic methodology for integrating CFD simulation and parameter sensitivity analysis, offering quantitative guidance for the hydraulic performance evaluation and design of underground pump station systems.

MLPDV-RSSI: Multi-Layer Perceptron-Based DV-Hop with Received Signal Strength Indicator for Localization in Wireless Sensor Networks

Scipedia content — Mon, 14 Sep 2026 12:10:06 +0200

In recent years, localization has emerged as a significant aspect of Wireless Sensor Networks (WSNs). The Distance Vector-Hop (DV-HOP) algorithm is a widely used localization approach for WSN due to its costeffectiveness and simplicity. However, it suffers from accuracy issues due to the random deployment of sensor nodes, and thus, the calculation errors in hop distance and the minimum number of hops. To address these challenges, this paper proposes Multi-Layer Perceptron-based DV-HOP with Received Signal Strength Indicator (MLPDV-RSSI), a novel intelligent algorithm that enhances localization by integrating DV-Hop with Multilayer Perceptron (MLP) and Received Signal Strength Indicator (RSSI). The proposed algorithm employs the RSSI to determine the location of sensors within a defined hop from the anchor nodes (ANs). Meanwhile, the DV-HOP method effectively localizes sensor nodes over multiple hops from ANs. Combining DV-HOP with the RSSI in the proposed model enhances the accuracy, particularly in dynamic indoor environments where measurements fluctuate. Feeding the RSSI measurements, hop count, and average hop distance values into the MLP for training while adjusting the weights reduces the error in the average hop distance and further improves accuracy. The simulation results show that the proposed method applied in a static environment achieves a significant accuracy improvement by up to 42% compared to the benchmarks. In a dynamic environment under 40% mobile nodes with non-Gaussian interference, a significant performance improvement by up to 48.9% is achieved compared to DV-HOP.

Intelligent Well Trajectory Design Based on Optimal Offset Well and RL-PSO

Scipedia content — Mon, 14 Sep 2026 12:16:03 +0200

Well trajectory design is a fundamental component of drilling design. Conventional well trajectory parameter selection relies heavily on expert experience, requiring repeated manual adjustment, resulting in low design efficiency. Traditional trajectory optimization algorithms require pre-defined parameter ranges and fail to fully leverage the offset well data. To address these limitations, this paper presents a proof-of-concept intelligent well trajectory design scheme that fully exploits the formation and offset well data by integrating reinforcement learning and optimization theory. An optimal offset well screening strategy and a reinforcement learning-based PSO (RL-PSO) algorithm are proposed to realize automatic trajectory design without manual parameter configuration. First, multiple features including formation similarities are encoded and quantified; the entropy weight method is applied to calculate the weight of each feature, and the optimal offset well is selected through a multi-dimensional comprehensive evaluation system. Second, based on the trajectory profile and parameter range derived from the optimal offset well, reinforcement learning is adopted to dynamically adjust the inertia weight and acceleration coefficients of the PSO algorithm for adaptive trajectory parameter optimization. Experimental results indicate that compared with traditional design methods, the proposed RL-PSO framework, driven by optimal offset well data can generate more economical well trajectories and improve design efficiency under the studied experimental conditions, providing a feasible intelligent solution for automated well trajectory design.