Network Structure Parameters Research Articles

Pneumonia detection on x-ray image using improved depthwise separable convolutional neural networks

A single neural network model cannot capture intricate and diverse features due to its ability to learn only a finite set of patterns from the data. Additionally, training and utilising a single model can be computationally demanding. Experts propose incorporating multiple neural network models to address these constraints to extract complementary attributes. Previous research has highlighted challenges network models face, including difficulties in effectively capturing highly detailed spatial features, redundancy in network structure parameters, and restricted generalisation capabilities. This study introduces an innovative neural network architecture that combines the Xception module with the inverse residue structure to tackle these issues. Considering this, the paper presents a model for detecting pneumonia in X-ray images employing an improved depthwise separable convolutional network. This network architecture integrates the inverse residual structure from the MobileNetV2 model, using the rectified linear unit (ReLU) non-linear activation function throughout the entire network. The experimental results show an impressive recognition rate with a test accuracy of 97.24% on the chest x-ray dataset. This method can extract more profound and abstract image features while mitigating overfitting issues and enhancing the network's generalisation capacity.

Structural parameter optimization of radial basis function neural network based on improved genetic algorithm and cost function model

This paper investigates the structural parameter optimization of RBF networks with the goal of economic control. The cost function and its implementation method are analyzed, and the cost function model of RBF neural network is established. The weights of RBF neural network are determined using recursive least squares method, then the crossover and mutation operators of genetic algorithm are improved and a new adaptive genetic algorithm is designed to implement the economic control of RBF neural network. The optimized network structure parameters are applied to the RBF neural network for simulation through function example, and the results obtained are compared with those of ordinary RBF network training. It shows that the method proposed in this paper is superior in both error and prediction values.

Robustness of LEO satellite hypernetworks under different network topologies and attack strategies

The development of low earth orbit (LEO) satellites has become a crucial and essential component with the advancement of digitalization. However, LEO satellites also face various security challenges within their networks, and the failure of certain satellites may trigger cascading failures. To gain a deeper understanding of cascading failure process in LEO satellite networks and enhance its robustness, this letter proposes a novel satellite network model by combining complex network and hypernetwork theories. By altering the network structure and model parameters, the robustness performance of the LEO satellite network under three types of attacks and two different load distribution methods is thoroughly investigated. The results indicate that satellite networks with varying structures are distinct and exhibit different robustness characteristics under specific attack types. Furthermore, the robustness of satellite networks in multiple attack scenarios can be significantly enhanced by implementing improved load allocation strategies. These findings not only provide a theoretical foundation for the design and optimization of satellite networks but also ignite new perspectives and ideas for future research on their robustness.

A pore network modeling approach to bridge void ratio‐dependent soil water retention and unsaturated hydraulic conductivity curves

AbstractIn geotechnical engineering, understanding the relationship between soil permeability and deformation is essential, particularly for applications like earth dams, where compaction‐induced permeability reduction is crucial for performance optimization. In unsaturated soils, soil moisture content significantly impacts hydraulic conductivity. Traditionally, changes in unsaturated hydraulic conductivity have been linked to soil void ratios. However, a pore network modeling perspective reveals the significance of structural parameters, such as pore and throat size distribution and pore coordination number. This study introduces a pore network model to estimate unsaturated hydraulic conductivity based on void ratio‐dependent soil‐water retention curves. It examines how soil deformation at varying stress levels affects structural parameters and the water phase continuity. The model shows strong potential in predicting unsaturated hydraulic conductivity across different stress levels, aligning well with experimental data and established equations. Notably, the aspect ratio and coordination number parameters are most affected by stress levels. The study also presents relationships to describe changes in pore network structural parameters with soil void ratio, which can be used to predict soil‐water retention curves and unsaturated hydraulic conductivity at various stress levels.

A Networked Meta-Population Epidemic Model with Population Flow and Its Application to the Prediction of the COVID-19 Pandemic.

This article addresses the crucial issues of how asymptomatic individuals and population movements influence the spread of epidemics. Specifically, a discrete-time networked Susceptible-Asymptomatic-Infected-Recovered (SAIR) model that integrates population flow is introduced to investigate the dynamics of epidemic transmission among individuals. In contrast to existing data-driven system identification approaches that identify the network structure or system parameters separately, a joint estimation framework is developed in this study. The joint framework incorporates historical measurements and enables the simultaneous estimation of transmission topology and epidemic factors. The use of the joint estimation scheme reduces the estimation error. The stability of equilibria and convergence behaviors of proposed dynamics are then analyzed. Furthermore, the sensitivity of the proposed model to population movements is evaluated in terms of the basic reproduction number. This article also rigorously investigates the effectiveness of non-pharmaceutical interventions via distributively controlling population flow in curbing virus transmission. It is found that the population flow control strategy reduces the number of infections during the epidemic.

Automatic Metric Search for Few-Shot Learning.

Few-shot learning (FSL) aims to learn a model that can identify unseen classes using only a few training samples from each class. Most of the existing FSL methods adopt a manually predefined metric function to measure the relationship between a sample and a class, which usually require tremendous efforts and domain knowledge. In contrast, we propose a novel model called automatic metric search (Auto-MS), in which an Auto-MS space is designed for automatically searching task-specific metric functions. This allows us to further develop a new searching strategy to facilitate automated FSL. More specifically, by incorporating the episode-training mechanism into the bilevel search strategy, the proposed search strategy can effectively optimize the network weights and structural parameters of the few-shot model. Extensive experiments on the miniImageNet and tieredImageNet datasets demonstrate that the proposed Auto-MS achieves superior performance in FSL problems.

Butterfly valve erosion prediction based on LSTM network

When valves and pipelines are used, erosion wear is a major concern. Erosion wear can lead to equipment downtime, material replacement, and other issues, as well as seal surface failure. The research delves into the phenomenon of erosion in butterfly valves, crucial components utilized across diverse industrial sectors. Erosion primarily affects the valve's disc and seat regions, leading to premature wear and compromised performance. To address this issue, the research employs computational fluid dynamics (CFD) and machine learning techniques, including long short-term memory (LSTM) and BP neural networks, to predict erosion rates under various conditions (different valve openings and particle diameters). Results indicate that erosion escalates with time and larger valve openings, highlighting the need for proactive maintenance strategies. The LSTM model demonstrates superior predictive capabilities compared to the BP neural network, offering valuable insights for improving butterfly valve design and operational efficiency in industrial settings. Furthermore, to seek a more efficient network configuration, the intelligent search capabilities of the particle swarm optimization (PSO) algorithm have been utilized to systematically explore the optimal network structure parameters. The prediction results highlight the advantages of LSTM in handling time series data, particularly in predicting erosion rates with complex dynamic characteristics.

The Inversion Method of Shale Gas Effective Fracture Network Volume Based on Flow Back Data—A Case Study of Southern Sichuan Basin Shale

Fracture network fracturing is pivotal for achieving the economical and efficient development of shale gas, with the connectivity among fracture networks playing a crucial role in reservoir stimulation effectiveness. However, flow back data that reflect fracture network connectivity information are often ignored, resulting in an inaccurate prediction of the effective fracture network volume (EFNV). The accurate calculation of the EFNV has become a key and difficult issue in the field of shale fracturing. For this reason, the accurate shale gas effective fracture network volume inversion method needs to be improved. Based on the flow back characteristics of fracturing fluids, a tree-shaped fractal fracture flow back mathematical model for inversion of EFNV was established and combined with fractal theory. A genetic algorithm workflow suitable for EFNV inversion of shale gas was constructed based on the flow back data after fracturing, and the fracture wells in southern Sichuan were used as an example to carry out the EFNV inversion. The reliability of the inversion model was verified by testing production, cumulative gas production, and microseismic results. The field application showed that the inversion method proposed in this paper can obtain tree-shaped fractal fracture network structure parameters, fracture system original pressure, matrix gas breakthrough pressure, fracture compressibility coefficient, reverse imbibition index, equivalent main fracture half length, and effective initial fracture volume (EIFV). The calculated results of the model belong to the same order of magnitude as those of the HD model and Alkouh model, and the model has stronger applicability. This research has important theoretical guiding significance and field application value for improving the accuracy of the EFNV calculation.

A Remaining Useful Life Prediction Method of Mechanical Equipment Based on Particle Swarm Optimization-Convolutional Neural Network-Bidirectional Long Short-Term Memory

In industry, forecast prediction and health management (PHM) is used to improve system reliability and efficiency. In PHM, remaining useful life (RUL) prediction plays a key role in preventing machine failures and reducing operating costs, especially for reliability requirements such as critical components in aviation as well as for costly equipment. With the development of deep learning techniques, many RUL prediction methods employ convolutional neural network (CNN) and long short-term memory (LSTM) networks and demonstrate superior performance. In this paper, a novel two-stream network based on a bidirectional long short-term memory neural network (BiLSTM) is proposed to establish a two-stage residual life prediction model for mechanical devices using CNN as the feature extractor and BiLSTM as the timing processor, and finally, a particle swarm optimization (PSO) algorithm is used to adjust and optimize the network structural parameters for the initial data. Under the condition of lack of professional knowledge, the adaptive extraction of the features of the data accumulated by the enterprise and the effective processing of a large amount of timing data are achieved. Comparing the prediction results with other models through examples, it shows that the model established in this paper significantly improves the accuracy and efficiency of equipment remaining life prediction.

Research on light weight intelligent identification method of coal and gangue

Accurate identification of coal and gangue is essential for clean and efficient use of coal. Existing target detection algorithms are ineffective in detecting small-target and overlapping gangue, and contain complex network structure and large parameter volume, which cannot meet the demand of real-time detection of edge devices. To address the above problems, a lightweight detection and identification approach of coal gangue based on improved YOLOv5s is proposed. The depth-separable convolutions are used to replace ordinary convolutions, and the C3 (Concentrated-Comprehensive Convolution Block) Ghost module is constructed to replace all C3 modules in the YOLOv5s to reduce model computation and parameters. The CA (Coordinate Attention) attention mechanism is introduced to strengthen the attention to the target to be detected, suppress irrelevant background interference, and improve the detection accuracy of the model. The Focal- EIOU (Focal and Efficient Intersection Over Union) loss function was introduced to replace the original CIOU. Extensive experiments substantiated the proposed approach can effectively and quickly detect the small-target and overlapping coal gangue accurately, and the mAP (mean Average Presicion) reaches 97.7%. Compared with the original YOLOv5s, the proposed approach reduces the number of parameters and the amount of computation by 48.5% and 43%, respectively, under the premise of maintaining the same detection accuracy.

Adaptive residual convolutional neural network for compressive strength prediction of energetic materials using SEM images

The microstructure of materials has a great influence on the macroscopic mechanical properties of materials, and the relationship between them is of great significance to the design of materials. With the development of artificial intelligence and deep learning, many researchers have used convolutional neural networks (CNN) to correlate microscopic images with material properties, and achieved good results. However, for different types of material performance prediction tasks and datasets of different sizes, researchers need to design a special CNN network structure, and the adjustment steps of network structure parameters are cumbersome and time consuming. In this paper, an adaptive residual convolutional neural network with variable depth and width is proposed. We take the prediction error of the test dataset of the specific material performance prediction task as the evaluation index, and adaptively adjusts the network structure based on simulated annealing algorithm to find the best network structure. Finally, it is verified on the dataset composed of scanning electron microscopy (SEM) images and compressive strength of polymer explosive materials. Mean absolute percentage error (MAPE) of our model is 1.3% and the goodness of fit (R2 ) is 0.943. Further, we use Grad-CAM to explore the correlation between material microstructure and material properties.

Comparison of threshold algorithms for automatic processing of fat crystal microscopic images based on ImageJ

AbstractMicroscopic image analysis is a crucial tool in fat crystallization research, enabling the analysis of crystal size, network structure, fractal dimension and other parameters through binarization. It is essential to seek an appropriate thresholding algorithm to binarize fat crystal images, which plays a vital role in image segmentation. In this article, the effectiveness of 17 thresholding algorithms such as Default, Mean, IsoData, Otsu, Li and Triangle were analyzed in processing fat crystal images with different shapes, background colors and image intensities. This was expected to discover a stable and objective thresholding algorithm for the binarization of fat crystal images. The performance evaluation was conducted according to the peak signal noise ratio (PSNR), structural similarity index (SSIM) and region non‐uniformity (RNU) parameter. Moreover, the comparative analysis of crystal size error, crystal area fraction and intraclass correlation coefficients (ICC) for fractal dimension values would provide a foundation for the selection of thresholding techniques for fat crystal network images. The results indicated that the Default algorithm exhibited remarkable robustness and applicability with high‐quality and stable outputs in fat crystal image processing.

Deep Reinforcement Learning Lane-Changing Decision Algorithm for Intelligent Vehicles Combining LSTM Trajectory Prediction

Intelligent decisions for autonomous lane-changing in vehicles have consistently been a focal point of research in the industry. Traditional lane-changing algorithms, which rely on predefined rules, are ill-suited for the complexities and variabilities of real-world road conditions. In this study, we propose an algorithm that leverages the deep deterministic policy gradient (DDPG) reinforcement learning, integrated with a long short-term memory (LSTM) trajectory prediction model, termed as LSTM-DDPG. In the proposed LSTM-DDPG model, the LSTM state module transforms the observed values from the observation module into a state representation, which then serves as a direct input to the DDPG actor network. Meanwhile, the LSTM prediction module translates the historical trajectory coordinates of nearby vehicles into a word-embedding vector via a fully connected layer, thus providing predicted trajectory information for surrounding vehicles. This integrated LSTM approach considers the potential influence of nearby vehicles on the lane-changing decisions of the subject vehicle. Furthermore, our study emphasizes the safety, efficiency, and comfort of the lane-changing process. Accordingly, we designed a reward and penalty function for the LSTM-DDPG algorithm and determined the optimal network structure parameters. The algorithm was then tested on a simulation platform built with MATLAB/Simulink. Our findings indicate that the LSTM-DDPG model offers a more realistic representation of traffic scenarios involving vehicle interactions. When compared to the traditional DDPG algorithm, the LSTM-DDPG achieved a 7.4% increase in average single-step rewards after normalization, underscoring its superior performance in enhancing lane-changing safety and efficiency. This research provides new ideas for advanced lane-changing decisions in autonomous vehicles.

Quantitative Detection of Defects in Multi-Layer Lightweight Composite Structures Using THz-TDS Based on a U-Net-BiLSTM Network.

Multi-layer lightweight composite structures are widely used in the field of aviation and aerospace during the processes of manufacturing and use, and, as such, they inevitably produce defects, damage, and other quality problems, creating the need for timely non-destructive testing procedures and the convenient repair or replacement of quality problems related to the material. When using terahertz non-destructive testing technology to detect defects in multi-layer lightweight composite materials, due to the complexity of their structure and defect types, there are many signal characteristics of terahertz waves propagating in the structures, and there is no obvious rule behind them, resulting in a large gap between the recognition results and the actual ones. In this study, we introduced a U-Net-BiLSTM network that combines the strengths of the U-Net and BiLSTM networks. The U-Net network extracts the spatial features of THz signals, while the BiLSTM network captures their temporal features. By optimizing the network structure and various parameters, we obtained a model tailored to THz spectroscopy data. This model was subsequently employed for the identification and quantitative analysis of defects in multi-layer lightweight composite structures using THz non-destructive testing. The proposed U-Net-BiLSTM network achieved an accuracy of 99.45% in typical defect identification, with a comprehensive F1 score of 99.43%, outperforming the CNN, ResNet, U-Net, and BiLSTM networks. By leveraging defect classification and thickness recognition, this study successfully reconstructed three-dimensional THz defect images, thereby realizing quantitative defect detection.

Many-objective coevolutionary learning algorithm with extreme learning machine auto-encoder for ensemble classifier of feedforward neural networks

In artificial neural network (ANN) learning, empirical risk can be expressed by training error, while structural risk can be expressed by network complexity. Learning from data is often seen as a tradeoff between these two risks. Additionally, balancing training error and validation error can overcome the overfitting problem to some extent. It is known that the network performance is also related to regularization term. In order to consider four factors, i.e. training error, validation error, network complexity and regularization term simultaneously in the training process of a single-hidden layer feedforword neural network (SLFN), a many-objective coevolutionary learning algorithm (MOCELA) integrated with extreme learning machine auto-encoder (ELMAE), called MOCELA-ELMAE is presented. In MOCELA, the non-dominated sorting genetic algorithm III (NSGA-III) is improved for handling the many-objective optimization model with hybrid variables, where binary coding is used for structure learning and real coding is utilized for representing input parameters, referring to all input weights and hidden biases of the AE network. Output parameters of AE, i.e. output weights are analytically calculated by the non-iterative learning rule. The network structure and connection parameters of SLFN are determined based on those of AE. A set of Pareto optimal solutions are eventually collected by the MOCELA-ELMAE, which represents multiple optimal SLFNs. To make the final decision, three best SLFNs with minimum validation errors are selected as the base classifiers for selective ensemble learning. Extensive experiments are implemented on the benchmark classification data sets of UCI machine learning repository, and obvious improvements have been observed when the proposed MOCELA-ELMAE is compared with the NSGA-III based on hybrid coding and completely non-iterative learning of SLFN respectively. The experimental results also illustrate that the MOCELA-ELMAE performs much better than other state-of-the-art learning algorithms on many data sets.

Evaluation of neural network models and quality forecasting based on process time‐series data

AbstractComplex industrial process modelling is critically important within the context of industrial intelligence. In recent years, soft sensor techniques based on neural networks have become increasingly popular for modelling nonlinear industrial processes. This paper proposes an integrated framework of neural network modelling and evaluation for nonlinear dynamic processes. This framework achieves an integrated solution for modelling, prediction, evaluation, and network structure parameter selection. It can be applied to noisy sensors and dense data in the time domain. The framework's proposed evaluation mechanism employs two novel evaluation metrics, the variational auto‐encoder (VAE)‐based Kullback–Leibler (KL) divergence metric and the maximum likelihood estimation‐based J metric, which both evaluate the model by mining the statistical properties of the residuals. The framework models the dynamic process with a model order based‐gated recurrent units (MOb‐GRU) neural network and a modified transformer model. Numerical experiments demonstrate that the evaluation mechanism functions properly in scenarios with multiple signal‐to‐noise ratios and multiple noise statistical properties and that the framework produces accurate modelling results.

Prediction of Thermal Conductance of Complex Networks with Deep Learning

Predicting thermal conductance of complex networks poses a formidable challenge in the field of materials science and engineering. This challenge arises due to the intricate interplay between the parameters of network structure and thermal conductance, encompassing connectivity, network topology, network geometry, node inhomogeneity, and others. Our understanding of how these parameters specifically influence heat transfer performance remains limited. Deep learning offers a promising approach for addressing such complex problems. We find that the well-established convolutional neural network models AlexNet can predict the thermal conductance of complex network efficiently. Our approach further optimizes the calculation efficiency by reducing the image recognition in consideration that the thermal transfer is inherently encoded within the Laplacian matrix. Intriguingly, our findings reveal that adopting a simpler convolutional neural network architecture can achieve a comparable prediction accuracy while requiring less computational time. This result facilitates a more efficient solution for predicting the thermal conductance of complex networks and serves as a reference for machine learning algorithm in related domains.

Neural Network Models for Predicting Solubility and Metabolism Class of Drugs in the Biopharmaceutics Drug Disposition Classification System (BDDCS).

The biopharmaceutics drug disposition classification system (BDDCS) categorizes drugs into four classes on the basis of their solubility and metabolism. This framework allows for the study of the pharmacokinetics of transporters and enzymatic metabolization on biopharmaceuticals, as well as drug-drug interactions in the body. The objective of the present study was to develop computational models by neural network models and structural parameters and physicochemical properties to estimate the class of a drug in the BDDCS system. In this study, deep learning methods were utilized to explore the potential of artificial and convolutional neural networks (ANNs and CNNs) in predicting the BDDCS class of 721 substances. The structural parameters and physicochemical properties [Abraham solvation parameters, octanol-water partition (log P) and over the pH range 1-7.5 (log D), number of rotatable bonds, hydrogen bond acceptor numbers, as well as hydrogen bond donor count] are calculated with various software. These compounds were then split into a training set consisting of 602 molecules and a test set of 119 compounds to validate the models. The results of this study showed that neural network models using applied parameters of the drug, i.e., log D and Abraham solvation parameters, are able to predict the class of solubility and metabolism in the BDDCS system with good accuracy. Neural network models are well equipped to deal with the relations between the structural parameters and physicochemical properties of drugs and BDDCS classes. In addition, log D is a more suitable parameter compared with log P in predicting BDDCS.

A Lightweight Unsupervised Intrusion Detection Model Based on Variational Auto-Encoder.

With the gradual integration of internet technology and the industrial control field, industrial control systems (ICSs) have begun to access public networks on a large scale. Attackers use these public network interfaces to launch frequent invasions of industrial control systems, thus resulting in equipment failure and downtime, production data leakage, and other serious harm. To ensure security, ICSs urgently need a mature intrusion detection mechanism. Most of the existing research on intrusion detection in ICSs focuses on improving the accuracy of intrusion detection, thereby ignoring the problem of limited equipment resources in industrial control environments, which makes it difficult to apply excellent intrusion detection algorithms in practice. In this study, we first use the spectral residual (SR) algorithm to process the data; we then propose the improved lightweight variational autoencoder (LVA) with autoregression to reconstruct the data, and we finally perform anomaly determination based on the permutation entropy (PE) algorithm. We construct a lightweight unsupervised intrusion detection model named LVA-SP. The model as a whole adopts a lightweight design with a simpler network structure and fewer parameters, which achieves a balance between the detection accuracy and the system resource overhead. Experimental results on the ICSs dataset show that our proposed LVA-SP model achieved an F1-score of 84.81% and has advantages in terms of time and memory overhead.

Synthesis and Performance Evaluation of a Novel High-Temperature-Resistant Thickener.

Successful exploitation of carbonate reservoirs relies on the acid-fracturing process, while the thickeners used in this process play a key role. It is a common engineering problem that thickeners usually fail to function when used in high-temperature environments. Until now, no research has ventured into the field of synthesizing thickeners which can be effectively used at ultra-high temperatures up to 180 °C. In our current study, a novel high-temperature-resistant polyacrylamide thickener named SYGT has been developed. The thermal gravimetric analysis (TGA) reveals that SYGT is capable of withstanding temperatures of up to 300 °C. Both our scanning electron microscopy (SEM) and rheological analysis demonstrate that the SYGT exhibits excellent resistance to both temperature and shear. At 180 °C, the viscosity of the SYGT aqueous solution is no lower than 61.7 mPa·s at a 20% H+ concentration or high salt concentration, and the fracture conductivity of the thickened acid reaches 6 D·cm. For the first time, the influence of the polymer spatial network's structural parameters on the viscosity of polymer solutions has been evaluated quantitatively. It was discovered that the length and surrounding area of the SNS skeleton have a synergistic effect on the viscosity of the polymer solution. Our experiments show that SYGT effectively reduces the acid-rock reaction rate and filtration loss under harsh working conditions such as high temperature, strong shear, high salinity, and a high concentration of acid. The synthesized acid-fracturing thickener (SYGT) has wide application potential in the development of carbonate reservoirs under high-temperature conditions.