https://cse.guilan.ac.ir/ Computational Sciences and Engineering en daily 1 Thu, 01 Jan 2026 00:00:00 +0330 Thu, 01 Jan 2026 00:00:00 +0330 https://cse.guilan.ac.ir/article_9332.html Inverse Partial Differential Equations (IPDEs) represent a class of ill-posed problems that frequently arise in engineering and applied sciences, such as heat conduction, diffusion, and imaging. Conventional approaches for solving IPDEs often rely on iterative or regularization techniques, which may suffer from instability and sensitivity to measurement errors. Recently, metaheuristic algorithms have emerged as powerful alternatives due to their global search ability and robustness. However, most of the existing studies, including those employing Teaching-Learning Based Optimization (TLBO), still face challenges in balancing accuracy, convergence speed, and computational cost. In this paper, we propose a novel hybrid metaheuristic algorithm combining Grey Wolf Optimizer (GWO) and Cuckoo Search (CS), referred to as HGWO-CS, for solving IPDEs. The hybrid strategy leverages the leadership hierarchy and hunting mechanism of GWO with the Lévy flight-based exploration of CS to achieve both efficient exploitation and diverse exploration. The method is applied to several benchmark IPDE problems, including inverse heat conduction problems, without requiring any prior assumption about the form of the unknown function. Numerical experiments demonstrate that the HGWO-CS approach achieves higher accuracy, faster convergence, and improved stability compared to existing algorithms such as TLBO, GA, and PSO. The main novelties of this work summarized as follows: the proposed method does not assume any predefined functional form for the unknown boundary conditions; the hybridization strategy effectively balances exploration and exploitation, resulting in improved accuracy, convergence speed, and robustness against noisy measurements. The proposed algorithm thus provides a promising tool for tackling complex IPDEs in engineering applications. https://cse.guilan.ac.ir/article_9333.html Given the importance of breast cancer detection and the increasing prevalence of this disease, along with its high annual mortality rate, extensive research has been conducted in recent years on medical image analysis for this purpose. In this paper, a multi-stage method is presented based on image segmentation of healthy and unhealthy (cancerous) tissues and a hybrid classification approach for determining the type of cancer (benign or malignant). In the proposed method, after noise reduction, an improved U-Net model is employed for image segmentation and detection of tumor candidate regions. For images identified as unhealthy, contour-based feature extraction is applied, followed by a hybrid ensemble classification method using majority voting among base classifiers to determine the cancer type. The proposed approach has been evaluated on a standard ultrasound image dataset consisting of healthy, benign, and malignant samples. The proposed method achieved a segmentation accuracy of 97.43% using the enhanced U-Net and an overall system accuracy of 94% for breast cancer diagnosis, outperforming other recent state-of-the-art techniques on the same dataset. https://cse.guilan.ac.ir/article_9430.html Heat exchangers are a major component of many heat-related systems, and improving their heat transfer efficiency is a significant engineering challenge. Though the problems of geometric optimization and vibration-assisted mechanisms have been widely discussed, little has been done to examine the impact of partly elastic intermediate walls and oscillatory behavior in counter-flow configurations. This paper numerically simulates the fluid flow and heat transfer in a counter-flow plate heat exchanger with partially elastic intermediate walls with a fully coupled fluid-structure interaction (FSI) methodology. The thermal and hydrodynamic performance of an elastic wall location, oscillation mode, frequency, and amplitude are analyzed. They indicate that the proposed configurations are highly effective in promoting heat transfer, where the increase in Nusselt number is between 23% and 83% in the optimal mid-uniform oscillation scenario. When the oscillation frequency changes to 3 Hz, the Nusselt number is increased by as much as 48 percent, indicating that oscillating elastic walls have a great potential in enhancing heat transfer. Considering the importance of the overall thermo-hydraulic performance (PEC), the performance evaluation criterion was also calculated, confirming that the optimal configurations provide a net performance benefit, with PEC values exceeding 1.0 and reaching up to 1.29. https://cse.guilan.ac.ir/article_9444.html In modern retail management, there is a high demand for being able to efficiently identify empty shelves. To address the inherent limitations of current monitoring systems in handling dense arrangements and geometrically diverse products, To address this image processing problem, we propose CoGY-Net—a robust, intelligent Out-of-Stock (OOS) detection framework that takes RGB images of retail shelves as input. Our approach significantly enhances the standard YOLO architecture through two primary innovations. First, we integrate the Contourlet Transform as a geometric pre-processor to improve the extraction of curved product features within cluttered backgrounds, leveraging its superior directionality over traditional transforms. Second, we employ the Golden Eagle Optimizer (GEO), a metaheuristic algorithm, to eliminate the inefficiency of manual tuning by autonomously identifying the ideal training hyperparameters and anchor boxes tailored to the specific dataset. Furthermore, to ensure the system remains reliable across varying shelf depths, we implemented a scale-invariant dynamic gap analysis logic to pinpoint empty spaces accurately. In this way we manage to fill the research gap which was the lack of an automated and geometry-aware detection framework capable of handling dense shelf layout and curved products in real environments. The system was evaluated on the Out-Of-Stock-23 dataset. Experimental results demonstrate that CoGY-Net achieves an accuracy 90% and provides a high-precision, automated solution with superior stability, making it highly suitable for seamless integration into real-time smart retail environments and autonomous inventory systems.  https://cse.guilan.ac.ir/article_9542.html Polynomial interpolation is a fundamental tool in numerical analysis, with its accuracy classically characterized by the Lagrange remainder term. This term involves an unknown mean value point , which depends on  and the function . Consequently, while the formula provides an exact error representation, its practical utility for a priori error estimation is limited, as determining a sharp, computable bound for high-order derivatives is often challenging. This paper introduces a novel probabilistic framework to address this longstanding limitation. Instead of treating  as an unknown deterministic value, we model its location probabilistically. By considering the interpolation nodes as random variables or analyzing the distribution of  for a fixed , we derive a statistical estimate for its expected value. This approach allows us to propose a specific, computable value for  that provides a highly accurate estimate of the actual interpolation error. The theoretical findings are substantiated with several numerical examples. These experiments demonstrate that our probabilistic estimate of the remainder term consistently aligns with the true error, offering a practical and powerful alternative to traditional worst-case error bounds. This method provides a new perspective on error analysis in approximation theory, bridging deterministic numerical methods with probabilistic techniques. https://cse.guilan.ac.ir/article_9445.html The Sardar Subequation method (SSM) is utilized in this paper to construct exact soliton solutions of the nonlinear Zoomeron equation (ZE). A feature of the SSM is its ability to produce multiple types of soliton solutions, including dark, bright, singular, periodic, and mixed dark–bright solitons. The findings indicate that the SSM is both efficient and powerful, while being computationally simple and extendable to other NLPDEs. To demonstrate the behavior and development of the obtained exact solutions, a variety of graphical representations is employed, including 3D, contour and 2D profiles. To depict these solutions, MATLAB software is employed, with different values assigned to the relevant parameters https://cse.guilan.ac.ir/article_9451.html Congestion remains a critical challenge in Software Defined Networks (SDNs), particularly in data center and Internet of Things (IoT) environments where high traffic dynamics and low latency requirements coexist. While SDN enables centralized traffic control, many existing congestion management approaches are either reactive—responding only after congestion occurs—or rely on computationally intensive learning-based prediction models that limit practical deployment. This paper proposes a lightweight and proactive congestion prediction and mitigation framework that tightly integrates real-time network monitoring, early congestion anticipation, and adaptive traffic rerouting. The approach predicts congestion using aggregated link utilization and delay metrics over sliding windows and proactively installs alternative OpenFlow forwarding rules based on optimized path selection using Dijkstra’s algorithm. Unlike complex predictive models, the proposed method requires no training data and incurs minimal controller overhead, making it suitable for real-world SDN deployments. The framework is implemented using the RYU controller and evaluated in a Mininet-based data center topology. Experimental results demonstrate significant improvements in throughput, packet loss, retransmission rate, and network efficiency compared to shortest-path routing and reactive update schemes. These results highlight the effectiveness of combining lightweight prediction with proactive control, offering a practical and scalable solution for congestion management in modern SDN environments. https://cse.guilan.ac.ir/article_9461.html This study presents an islanding detection strategy for inverter interfaced distributed generation wherein the detection is governed by a learning-based characterization of load. In contrast to conventional frequency-based relays, the proposed approach deliberately introduces a controlled reactive power imbalance to induce a measurable frequency deviation while adaptively tuning its response according to inherent load attributes, including the resonant frequency, the quality factor, and robustness against non-Gaussian load achieved through Gaussian Model (GMM) clustering. To identify these characteristics, load signatures are extracted and processed within a hybrid machine-learning framework, are employed to cluster operating conditions into representative groups, and a regression estimator is applied to accurately infer the corresponding load coefficients. Based on these features, an optimal d-q axis current modulation scheme has been formulated to ensure distinct frequency deviations under islanded conditions. The effectiveness of the proposed methodology has been evaluated across a broad range of load scenarios, including those compliant with IEEE 1547 standards. Simulation results demonstrate that the method consistently detects islanding within 31 ms, while significantly reducing the non-detection zone compared to widely adopted Q-f droop and adaptive reactive power control techniques. Moreover, the proposed scheme alleviates transient voltage and frequency disturbances during grid disconnection, enabling smoother operational transitions. By integrating data-driven load assessment with optimal tuning of control parameters, the proposed framework enhances system reliability and detection responsiveness without requiring additional sensing hardware. Consequently, this approach serves as a promising solution for the robust and safe integration of inverter-based renewable energy resources in modern distribution networks. https://cse.guilan.ac.ir/article_9475.html Direct Numerical Simulation (DNS) of turbulent flows is computationally prohibitive for most practical engineering applications. Instead, reduced-order models such as Reynolds-Averaged Navier–Stokes (RANS) equations are widely used, but they suffer from inherent structural errors, particularly in flows with separation and adverse pressure gradients. Quantifying the uncertainty of these models is essential for assessing the reliability of their predictions. This study presents a physics-informed deep learning framework for the direct quantification of structural model-form uncertainty in RANS simulations. The proposed approach employs a deep neural network (specifically, a Physics-Informed Neural Network—PINN) trained on high-fidelity wavy-wall flow data to learn the discrepancy between the RANS-predicted and reference Reynolds stress tensors in the form of perturbations to the eigenvalues and eigenvectors of the turbulence anisotropy tensor. Unlike common methods that focus on parametric uncertainty, this technique corrects the intrinsic model form error while preserving the physical realizability constraints of the Reynolds stress. The trained model is then rigorously validated on two geometrically and physically distinct, unseen flow configurations: an asymmetric planar diffuser and a periodic hill. Results demonstrate that the neural-network-based model significantly improves the accuracy of baseline RANS predictions. This improvement is evident in velocity profiles, turbulent kinetic energy, shear stress, and, most notably, in the prediction of flow separation and lift and drag coefficients. This work establishes that learning perturbations in eigenspace provides an effective, generalizable, and physics-constrained approach for assessing and reducing model uncertainty in engineering turbulence simulations. https://cse.guilan.ac.ir/article_9538.html Early detection and diagnosis of breast cancer masses can significantly reduce mortality associated with this disease. This paper introduces a system designed to identify cancerous masses in mammography images. The proposed system is composed of four phases: pre-processing, feature extraction, feature selection, and classification. In the pre-processing phase, the region of interest is isolated from the image, noise is eliminated using a median filter, and contrast is enhanced through histogram adjustment. In the feature extraction phase, the features pertinent to shape, histogram, and texture are extracted from the region of interest. In the third phase, the dual-objective coral reef algorithm is employed for feature selection. Finally, an SVM classifier is utilized in the classification phase. The proposed method was applied to 100 images from the MIAS database and 300 images from the CBIS-DDSM database. Based on the quantitative results, the system demonstrated promising performance in reducing classification errors. its accuracy, sensitivity, specificity, AUC, MCC, and F1-Score values for the MIAS database were calculated as 95.4%, 99.18%, 91.79%, 0.91, 0.95, and 95.32%, respectively, and for the CBIS-DDSM dataset database were calculated as 96.73%, 98%, 95.46%, 0.934, 0.968, and 96.71%, respectively. https://cse.guilan.ac.ir/article_9566.html The primary purpose of this study is to enhance stock price forecasting accuracy by jointly modeling numerical market data, semantic information from financial news, investor sentiment, and structural relationships among stocks. Addressing the limitations of traditional price-based and standalone deep learning models, this research focuses on uncovering hidden nonlinear and relational dynamics in an emerging stock market. This study proposes an integrated framework that combines a Stock Knowledge Graph (SKG), sentiment analysis, and a hybrid deep learning architecture. Structural and semantic relationships among stocks, investors, and financial entities are modeled using Node2Vec and TransE embeddings. These features are fused with technical indicators, financial news sentiment, and user-generated sentiment, and processed through a CNN–LSTM–Attention model. The framework is empirically evaluated using 4,319 trading days of data from the Iranian stock market, with multiple benchmark models for comparison. Empirical results demonstrate that the proposed model significantly outperforms baseline CNN, LSTM, and Transformer models across multiple evaluation metrics. The hybrid CNN–LSTM–Attention architecture achieves the lowest RMSE (0.754) and the highest explanatory power (R² = 0.91). Incorporating sentiment and semantic features reduces prediction error by more than 28% compared with models relying solely on technical indicators. Knowledge graph embeddings effectively capture latent inter-stock and semantic relationships, further improving forecasting performance. The findings confirm that integrating structural, semantic, and behavioral information within a unified deep learning framework substantially enhances stock price prediction accuracy. https://cse.guilan.ac.ir/article_9567.html This paper has discussed the selection of optimal increasing censorship plan based on Pittman Proximity Criterion. For small sample sizes, the Pittman proximity probabilities were clearly calculated and also it has been indicated that optimal increasing censorship plan is the censorship plan from right of regular second type. It has been inferred that this could be the same in all sample sizes (small and great). There has been also provided an algorithm for numerical calculations for probabilities of Pittman approximation between both growing censorship plans of sample size. The methodology is extended to capital market applications, including credit risk modeling and option pricing. https://cse.guilan.ac.ir/article_9568.html This study numerically investigates mixing performance in an active micromixer combining geometric modification and oscillating wall excitation. The proposed design features two variable-angle inlets (45°, 60°, 90°, and 180°), a circular mixing chamber (R = 10 and 12.5 mm), and a flexible oscillating outlet section (2-8 Hz, 1-3.5 mm amplitude). A two-way fluid-structure interaction (FSI) approach in COMSOL Multiphysics captures the mutual effects between fluid flow and the elastic wall. Water enters at 20°C and 80°C with Reynolds numbers of 100-825 in laminar flow regime, with average outlet temperature serving as the mixing indicator. Results show that increasing oscillation frequency and amplitude enhances mixing, with 8 Hz and 3 mm identified as optimal. The 60° Y-configuration outperforms the conventional T-mixer (90°) and parallel inlet (180°). While the mixing chamber generally improves performance, its effect is less pronounced at the optimal 60° angle. Unequal inlet velocities generate Kelvin-Helmholtz instability, creating shear-induced vortices that further enhance mixing. The optimal configuration—60° inlet angle, 10 mm chamber radius, and 8 Hz, 3 mm wall oscillation—achieves a 92% mixing index at Re = 825, a 58% improvement over the base T-mixer. Estimated Nusselt numbers range from 9.2 to 13.8, with friction coefficients of 0.42-0.52. All configurations reach steady mixing within 6 seconds. These findings provide valuable insights for designing high-efficiency active micromixers for microfluidic applications requiring rapid homogenization of non-isothermal fluids. https://cse.guilan.ac.ir/article_9569.html The rapid spread of fake news on digital platforms poses a significant threat to informed public discourse and societal trust. While Large Language Models (LLMs) like BERT have shown remarkable accuracy in automated fake news detection, their opaque nature hinders user trust and understanding. This paper presents a framework that combines the high predictive performance of BERT with post-hoc interpretability techniques to enhance both the effectiveness and transparency of fake news detection systems. Specifically, we fine-tune BERT for binary fake news classification on the COVID-19 Fake News Dataset and employ Local Interpretable Model-agnostic Explanations (LIME) and BERT attention visualization to elucidate the model's decision-making process. Our results demonstrate that the fine-tuned BERT model achieves excellent performance, with an accuracy of 97.66% and an F1-score of 97.49% on the test set. Furthermore, LIME explanations highlight the contribution of specific words to individual predictions, while attention visualizations reveal which token relationships the model deems important. This integrated approach underscores that "truth" in machine prediction encompasses not only high accuracy but also explainability, thereby fostering greater confidence in automated fake news detection systems. https://cse.guilan.ac.ir/article_9579.html Inverse Heat Conduction Problems (IHCPs) are highly ill-posed, sensitive to measurement noise, and computationally demanding. This study proposes a hybrid SMA–PSO algorithm combining the exploratory strength of the Slime Mould Algorithm (SMA) with the fast convergence of Particle Swarm Optimization (PSO). In the proposed framework, SMA guides the early search to maintain diversity, while PSO progressively refines promising solutions, effectively balancing global exploration and local exploitation. The inverse problem is formulated as an unconstrained optimization task to reconstruct unknown boundary heat fluxes by minimizing discrepancies between simulated and observed temperatures. The algorithm is tested on standard 1D and 2D IHCP benchmarks under noise-free and noisy conditions (up to 7% Gaussian noise). Results from 30 independent runs show that SMA–PSO outperforms standalone SMA, PSO, and the GA–PSO hybrid. For a 2D case with 3% noise, SMA–PSO reduces RMSE by 27% versus PSO, 39% versus SMA, and accelerates convergence by 18% compared to GA–PSO. Paired t-tests confirm statistical significance. The method demonstrates strong noise robustness, stable convergence with low variance, and computational efficiency comparable to PSO. These findings highlight SMA–PSO as a reliable, efficient, and effective approach for solving complex IHCPs in engineering applications. https://cse.guilan.ac.ir/article_9583.html The main objective of this research is to estimate and simulate the upper and lower bounds of Asian option prices using the MCMC method and to evaluate the efficiency of this approach in improving pricing models and contributing to the development of financial markets. Therefore, this study falls within the field of numerical analysis and computational finance. In this research, the pricing bounds of Asian options are determined using the MCMC simulation method. The statistical population consists of possible price paths of the underlying asset, which are modeled under a specified stochastic process, such as geometric Brownian motion, and the resulting outcomes are analyzed. In this context, explicit formulas are provided for several Lévy models, including the Kou model, the Merton model, and the NIG model. In the next stage, the results obtained from each of the Kou, Merton, and Normal Inverse Gaussian models are compared with the results of the MC simulation. The methods applied in this study include the upper bound method and the lower bound method. The results indicate that the lower bound is faster and more accurate than the upper bound. In particular, the numerical analysis shows that the lower bound introduced in this study can be computed very quickly and provides a high level of accuracy. Furthermore, for all considered models, the upper bound is computed at a reasonably fast rate; however, it is slower than the lower bound, which involves a single integral. https://cse.guilan.ac.ir/article_9595.html Numerical study of the flow and residence time distribution of a microreactor with five parallel microchannels was considered. Three types were used in this work: type A, type B and type C. In type A and type C, trapezoidal structures were used for manifolds, while, in type B, the rectangular structure of manifold was used. The effect of manifold structure and the direction of inlet flow respect to the direction of the flow in microchannels were investigated on the residence time distribution. Along this considering, uniformity of flow distribution in microchannels was studied. Results show with increasing the velocity, the dimensionless variance increases. In the lower velocity, there isn’t a clear difference between type A and type C in the dimensionless variance value, but, with increasing the velocity, type C has lower dimensionless variance. So, flow in type C is more uniform than type A. while, the dimensionless variance for type B is less than type C. Also, in all types, type A, type B and type C, the tailing factor value is more than one, so, these types aren’t ideal model reactors. Based on another result, the channelization of flow was shown in the higher velocity. https://cse.guilan.ac.ir/article_9783.html The adoption of deep learning in healthcare for disease diagnosis is hindered by limited data availability due to strict patient privacy regulations. Federated learning (FL) enables collaborative training without raw data sharing, yet remains vulnerable to reconstruction attacks such as GAN-based inversion, especially from honest-but-curious central servers. Existing defenses like differential privacy trade model utility for privacy, while encryption adds heavy overhead. This paper proposes FedPrEx, a decentralized FL framework with pruning and exchange that eliminates the central aggregator. FedPrEx introduces three innovations: (1) randomized pruning where each client randomly drops 20% of model parameters before transmission to disrupt gradient inversion; (2) secure peer-to-peer exchange among edge servers using symmetric encryption and one time random pairing; and (3) hierarchical decentralized aggregation that prevents any single entity from accessing consecutive updates. We evaluate FedPrEx on MNIST and eICU datasets. Compared to the state of the art Fed Select framework, FedPrEx achieves higher accuracy (98.3% vs. 96.1% on MNIST) and more stable convergence without oscillations. Unlike differential privacy, FedPrEx imposes low computational overhead via pruning instead of noise addition, while resisting GAN based and collusion attacks through randomized exchange and periodic central model evaluation. Security analysis shows that the decentralized architecture with encrypted peer exchange and dynamic participation effectively mitigates internal and external privacy threats. FedPrEx offers a practical, scalable, privacy preserving solution for training deep learning models on distributed healthcare data without compromising model quality. https://cse.guilan.ac.ir/article_9784.html This study introduces a hybrid deep learning approach for medical image edge detection, integrating the SREM–SqueezeNet architecture with mathematical optimization techniques to enhance accuracy and computational efficiency. The proposed framework employs the lightweight and parameter-efficient structure of SqueezeNet, which enables high-performance edge extraction while maintaining a compact model suitable for deployment on resource-constrained medical systems. The research emphasizes a mathematical formulation of the convolutional neural networks optimization process, incorporating evaluation metrics such as entropy, precision, recall, F-measure, true positive rate, and accuracy to quantitatively assess edge detection quality. Experimental results demonstrate the superiority of the proposed method, achieving a significant reduction in entropy to 0.1153 and an improvement in the F-measure to 0.9154, outperforming conventional edge detection techniques. These outcomes highlight the potential of the mathematically optimized SREM–SqueezeNet hybrid model as an effective and reliable solution for medical image analysis, contributing to improved diagnostic precision and automated disease detection in clinical applications https://cse.guilan.ac.ir/article_9817.html Standard semidefinite programming (SDP) relaxations for the Minimum Sum-of-Squares Clustering problem often produce weak lower bounds. We tighten the relaxation by adding transitivity constraints that enforce the equivalence-relation structure of cluster assignments: if points i and j are together, and i and k are together, then j and k must be together as well. We prove that these constraints are valid inequalities for the original problem, and that the new relaxation dominates the standard one in bound quality. Numerical experiments on Ruspini, Iris, Wine, and synthetic datasets show that our model consistently outperforms both the standard SDP relaxation and a recent MILP formulation, with improvements reaching up to 49% on synthetic instances and over 20% on Wine. The gains are largest when clusters overlap or when the number of clusters or data size grows. This comes at a higher computational cost due to the O(n^3 ) additional constraints, but the trade-off is acceptable for moderate-scale problems. The proposed relaxation offers a practical way to strengthen SDP-based clustering by encoding logical assignment consistency directly within the convex framework.