Network Parameters Research Articles

Urban flood prediction using ensemble artificial neural network: an investigation on improving model uncertainty

Reducing the impact of artificial neural networks (ANN) affected by sources of uncertainty is crucial to improving the reliability of the flood prediction model. This study proposes an ensemble artificial neural network (EANN) model to predict the degree of flooding in coastal cities. Combined methods are used to reduce the model’s uncertainty, heuristic neural pathway strength feature selection is used to select inputs, the coupling method is used to optimize network architecture and parameters, and the integration method which paralleling three ANN models with different predicted lead periods ensemble together is used to capture output uncertainty. The EANN model has successfully predicted flooding in the Chinese coastal city of Macao during a typhoon, with convincing accuracy. The study also analyzed the impacts of both long and short training datasets with appropriate time intervals on ANN modeling performance. It was found that the performance of short training datasets, with appropriate time intervals, was similar to or better than models with long training datasets.

Research on the theory and method of reduced-hole blasting for large cross-section tunnel based on explosive energy dissipation

The traditional tunnel drilling and blasting method places cut holes at the lower center of the excavation face, resulting in an excessive number of blasting holes. With the continuous increase in cross-section area, this design concept can no longer meet the requirements of safe and efficient tunnel boring for large cross-section tunnels. This paper puts forward the theory and method of reduced-hole blasting for large cross-section tunnels, as an alternative to the traditional drilling and blasting method of the “more holes, less charge” design concept. Based on the explosion energy dissipation law and rock’s critical crushing energy dissipation characteristics, the calculation method of the extrapolation distance of the wedge-cut holes is given. The optimum extrapolation distance of the wedge-cut holes was verified using numerical simulation and field tests. The results show that the number of drilling holes can be reduced by about 15.8% using the theory and method proposed in this paper, and at the same time, the damage of retained rock can be effectively controlled. The results of this study can provide a reference for the design of blast network parameters for similar large cross-section tunnels.

Investigation of SnO2-based Thick Film Gas Sensors through Optimization of Artificial Neural Networks Parameters for Hazardous Gas Detection

In the present study, a systematic approach for optimizing artificial neural network parameters for the detection of a greenhouse gas carbon di-oxide (CO2) through SnO2-based thick film gas sensors for a wide range of concentrations is presented. The un-doped and Pd-doped SnO2 thick film is used as the sensor. The approach includes dataset preparation, input and output variable selection, artificial neural network architecture selection, training, validation, hyperparameter tuning, and testing. The input variables were selected based on their relevance to the problem at hand, such as the concentrations of the gas. The output variables were the changes in resistance of the sensor in response to the corresponding input variables. The artificial neural network architecture was carefully chosen, considering factors like the number of layers, neurons per layer, activation functions, and learning rate. Using the prepared dataset, the artificial neural network was trained by adjusting the connection weights between neurons to minimize the disparity between actual and predicted sensor responses. After training, the accuracy of the artificial neural network and generalization ability were assessed using a separate dataset for validation. The approach in the presented study can be used for different types of gases, and can be utilized in various applications

Study on covert rate in the D2D networks with multiple non-colluding wardens

This paper investigates the covert communication of cellular link in the device-to-device (D2D)-enabled underlaying cellular network, including a base station, a cellular user, a D2D pair, and multiple non-colluding wardens. To conduct covert communication between the base station and the cellular user without being detected by the wardens, the underlaid D2D pair transmits with a random power to add uncertainty to wardens. We first model the detection of multiple non-colluding wardens, and the average minimum detection error probability of the worst case is derived. To measure the covert performance of the cellular link, the average covert rate of the cellular link and the outage probability of the D2D link are explored. Then, we formulate an optimization problem of maximizing the average covert rate under the constraint of the average minimum detection error probability, and an optimization algorithm is designed to solve the optimization problem to identify the optimal transmit power of the D2D pair. Last, the extensive numerical results are presented to show the impacts of network parameters on the covert performance, which indicates that although D2D communication causes interference to the cellular link, it can also help achieve covert communication by carefully designing its maximum transmit power.

Dynamic economic dispatch of integrated energy system based on generative adversarial imitation learning

Within the context of energy structure transformation, the inherent uncertainty of new energy sources presents severe challenges to the optimized operation of integrated energy system. Model-driven optimization methods are limited by the accuracy and complexity of models, resulting in lower solving efficiency. Meanwhile, deep reinforcement learning methods that have gained attention in recent years face difficulties in designing rewards and slow convergence due to dealing with high-dimensional and complex dynamic information of integrated energy system. Therefore, inspired by imitation learning, this paper proposes a real-time dispatching method for integrated energy system based on generative adversarial imitation learning. Firstly, a generator network model is designed within the deep reinforcement learning framework, and expert prior knowledge is introduced to assist the generator network in iterative learning, enhancing the model's convergence effect. Secondly, drawing on the generative adversarial concept, a discriminator network model is developed to recognize expert and generated strategies, further designing game rewards to assist in updating the generator network parameters and avoiding the impact of subjectively defined reward functions on dispatching decisions. Finally, simulation analysis demonstrates that the proposed method can calculate the operation optimization solutions for integrated energy system more efficiently, significantly improving the accuracy of dispatching decisions and the model's convergence efficiency.

Analysis of connection fault and service maintenance strategy for subsea horizontal clamp connector

The deep-sea environment restricts the collection of reliable data for subsea connectors, leading to a lack of available failure data to develop connection fault detection processes and maintenance strategies. Therefore, this paper proposes an FMECA-BN model. The model developed an algorithm for mapping Risk Priority Number (RPN) to Bayesian network (BN) parameters and enhanced the details for mapping BN structures to match the diversity of FMEA. This enables the establishment of a BN based on a given Failure Mode, Effects, and Criticality Analysis (FMECA). The model was applied to the development of the Subsea Horizontal Clamp Connector (SHCC) to quantify and evaluate failure modes during the connection stage, creating an effective detection process. Then, the system's reliability variations caused by component degradation were analyzed using transformed dynamic BN, offering information for maintenance strategies based on availability. The reliability analysis results of the FMECA-BN model demonstrate a statistical error of 4.5% compared to actual data samples, proving the model's efficacy in addressing issues of information asymmetry, incomplete reliability data, and uncertainty. The analysis revealed that the developed SHCC requires a maintenance cycle of 14,892 h to maintain a 0.99 reliability level, with a cumulative failure probability of 4.6% after 20 years.

Research on a Personalized Decision Control Algorithm for Autonomous Vehicles Based on the Reinforcement Learning from Human Feedback Strategy

To address the shortcomings of previous autonomous decision models, which often overlook the personalized features of users, this paper proposes a personalized decision control algorithm for autonomous vehicles based on RLHF (reinforcement learning from human feedback). The algorithm combines two reinforcement learning approaches, DDPG (Deep Deterministic Policy Gradient) and PPO (proximal policy optimization), and divides the control scheme into three phases including pre-training, human evaluation, and parameter optimization. During the pre-training phase, an agent is trained using the DDPG algorithm. In the human evaluation phase, different trajectories generated by the DDPG-trained agent are scored by individuals with different styles, and the respective reward models are trained based on the trajectories. In the parameter optimization phase, the network parameters are updated using the PPO algorithm and the reward values given by the reward model to achieve personalized autonomous vehicle control. To validate the control algorithm designed in this paper, a simulation scenario was built using CARLA_0.9.13 software. The results demonstrate that the proposed algorithm can provide personalized decision control solutions for different styles of people, satisfying human needs while ensuring safety.

Artificial intelligence model for predicting the load-bearing capacity of eccentrically compressed short concrete filled steel tubular columns

The purpose of this work is to develop the artificial neural network (ANN) model to determine the load-bearing capacity of concrete filled steel tubular (CFST) columns of circular cross-section in a wide range of input parameters. Short columns are considered for which deflections do not lead to a significant increase in the eccentricity of the axial force. The input parameters of the artificial neural network are the outer diameter of the pipe, the wall thickness, the yield strength of steel, the compressive strength of concrete, and the relative eccentricity of the axial force. The artificial neural network is trained on the synthetic data. For training, the dataset of 179,025 numerical experiments with different values of input parameters was generated. Numerical experiments were carried out using the finite element method in a simplified formulation, which makes it possible to reduce the three-dimensional problem of determining the stress-strain state of a CFST column to a two-dimensional problem. The results of testing the developed model on the data from full-scale experiments are presented.

Classification of Motor Imagery Tasks Derived from Unilateral Upper Limb based on a Weight-optimized Learning Model.

The accuracy of decoding fine motor imagery (MI) tasks remains relatively low due to the dense distribution of active areas in the cerebral cortex. To enhance the decoding of unilateral fine MI activity in the brain, a weight-optimized EEGNet model is introduced that recognizes six types of MI for the right upper limb, namely elbow flexion/extension, wrist pronation/supination and hand opening/grasping. The model is trained with augmented electroencephalography (EEG) data to learn deep features for MI classification. To address the sensitivity issue of the initial model weights to classification performance, a genetic algorithm (GA) is employed to determine the convolution kernel parameters for each layer of the EEGNet network, followed by optimization of the network weights through backpropagation. The algorithm's performance on the three joint classification is validated through experiment, achieving an average accuracy of 87.97%. The binary classification recognition rates for elbow joint, wrist joint, and hand joint are respectively 93.92%, 90.2%, and 94.64%. Thus, the product of the two-step accuracy value is obtained as the overall capability to distinguish the six types of MI, reaching an average accuracy of 81.74%. Compared to commonly used neural networks and traditional algorithms, the proposed method outperforms and significantly reduces the average error of different subjects. Overall, this algorithm effectively addresses the sensitivity of network parameters to initial weights, enhances algorithm robustness and improves the overall performance of MI task classification. Moreover, the method is applicable to other EEG classification tasks; for example, emotion and object recognition.

Dependency-aware deep generative models for multitasking analysis of spatial omics data.

Spatially resolved transcriptomics (SRT) technologies have significantly advanced biomedical research, but their data analysis remains challenging due to the discrete nature of the data and the high levels of noise, compounded by complex spatial dependencies. Here, we propose spaVAE, a dependency-aware, deep generative spatial variational autoencoder model that probabilistically characterizes count data while capturing spatial correlations. spaVAE introduces a hybrid embedding combining a Gaussian process prior with a Gaussian prior to explicitly capture spatial correlations among spots. It then optimizes the parameters of deep neural networks to approximate the distributions underlying the SRT data. With the approximated distributions, spaVAE can contribute to several analytical tasks that are essential for SRT data analysis, including dimensionality reduction, visualization, clustering, batch integration, denoising, differential expression, spatial interpolation, resolution enhancement and identification of spatially variable genes. Moreover, we have extended spaVAE to spaPeakVAE and spaMultiVAE to characterize spatial ATAC-seq (assay for transposase-accessible chromatin using sequencing) data and spatial multi-omics data, respectively.

Neural adaptive dynamic surface control for position tracking of electrohydraulic servo systems based on extended state observer

High-performance motion tracking of electro-hydraulic servo systems plays an important role in developing industrial technology. However, the system itself has parameter uncertainty, uncertain nonlinearity, and measurement noise, significantly decreasing the performance of conventional controllers designed based on nominal models. To improve the tracking performance of electro-hydraulic servo systems, an adaptive dynamic surface control method combining extended state observer and radial basis function neural network is proposed. Firstly, an extended state observer combining neural network output and parameter adaptation technology is designed without the need for speed and pressure feedback to observe the system’s unknown state variables and nonlinear matched disturbance. The design of neural networks is used to approximate mismatched disturbance with significant nonlinearity online. Then, the proposed extended state observer is introduced into the neural adaptive backstepping design, effectively compensating for simultaneous matched and mismatched disturbances, thereby suppressing most of the uncertain nonlinearity of the system. Meanwhile, online update of uncertain parameters further improves the tracking performance of the controller. In addition, the introduction of dynamic surface overcomes the “explosion of complexity” issue in backstepping design. The rigorous stability of the closed-loop system is proved by the Lyapunov stability theory. Finally, the effectiveness of the proposed controller is validated through comparative experiments on valve-controlled hydraulic cylinder.

APSSF: Adaptive CNN Pruning Based on Structural Similarity of Filters

Convolutional neural network (CNN) pruning is a technique used to remove redundant parameters from the network. By doing so, it aims to greatly reduce the computational complexity and scale of the network while still preserving its accuracy. In the CNN, the majority of parameters are weights that form filters. When it comes to pruning, it is more effective to focus on removing redundant filters rather than insignificant weights within filters. The essence of filter pruning lies in determining the significance or contribution of each filter. Filters that have a significant contribution are kept, while others are pruned. Current methods for calculating contribution in pruning often rely on weight magnitude or filter similarity. However, approaches based solely on assume that small weights are unimportant and ignore correlation between filters, which leads to a significant loss of network accuracy. Those based on filter similarity flatten filter tensors into a vector when calculating filter similarity, and lose the important structural information of filters, or the superposition information of the weight convolution in the corresponding space position. These limitations can compromise the accuracy and effectiveness of the pruning process. This paper proposes an adaptive CNN pruning method based on the structural similarity of filters (APSSF) by taking both the structural characteristics of and the correlation between filters into the consideration for pruning filters. APSSF efficiently calculates the distance between the filters by factoring in information from all the dimensions of filters, and clusters the filters according to the distance threshold determined adaptively according to the compression rate, and deletes a certain number of filters from each category. On the CIFAR10 and ImageNet datasets, APSSF outperforms several state-of-the-art methods. On the CIFAR100, APSSF reduces parameters of networks by 91.71% and 74.80% on VGG-16 and ResNet-34, respectively. The accuracy was decreased only by 0.03 on VGG-16, while on ResNet-34, it was increased by 0.04.

Spectral Ground Motion Models for Himalayas Using Transfer Learning Technique

ABSTRACT Predicting robust earthquake spectra is challenging, especially for data sparse regions such as India. Often, alternatives to the traditional data-driven regression analysis are used to develop empirical models for such regions. Advancing these efforts, the present study aims at exploring an alternative machine learning technique called Transfer learning, wherein a non-parametric deep neural network is trained for response (Sa) and Fourier spectra (FAS) of Himalayas, which uses network parameters that were derived for a large comprehensive database (NGA-West2). While the FAS is derived using magnitude, distance, focal depth, and site class, the Sa is scaled using FAS and significant duration.

Big data processing and analysis platform based on deep neural network model

Users are increasingly turning to big data processing systems to extract valuable information from massive datasets as the field of big data grows. Data analytics platforms are used by e-commerce enterprises to improve product suggestions and model business processes. In order to meet the needs of large-scale data center operation and maintenance management, Internet companies often use Flink to process log data. This paper takes the big data processing and analysis platforms built by Internet financial companies and large banks as examples, and implants a stock prediction model based on Deep Neural Network (DNN). In this context, this paper completes the following work: 1) The research status of big data processing and analysis platforms at home and abroad is introduced. 2) Drawing on the modular design idea, the commercial bank big data platform is designed and the functions of each sub-module are introduced. Then the basic principle and structure of Convolutional Neural Networks (CNN) are expounded. 3) The optimal parameters of Convolutional Neural Networks are selected through experiments, and then the trained model is used for experiments. It can be seen that the stock prediction model proposed in this article has a higher prediction accuracy compared to existing models, which also verifies the validity of the proposed model. Input the data and compare the obtained results with the actual results, and finally show that the model in this paper has a good performance on stock prediction.

Adaptive fusion transfer learning-based digital multitwin-assised intelligent fault diagnosis

With the development of digital twin technology, mechanical failures can be more comprehensively described through the interaction between sensors and twin models, which thereby demonstrate significant potential in fault diagnosis. However, due to the limitations of modeling methods, digital twin model construction poses challenges and exhibits poor generalization, making it difficult to apply across domains. To address these issues, this paper proposes a digital multitwin fusion diagnostic algorithm based on transfer learning. The digital multitwin comprise the source domain mechanistic twin, the target domain distributional feature 1D Auxiliary Classifier GAN (1D-ACGAN) twin, and the target domain time-series feature shared weight stacking LSTM (SWSLSTM) twin. First, based on the fault features collected by multiple sensors in the mechanism twin, a multichannel, multiscale mid-fusion diagnostic network is constructed. Second, the network is hierarchically transferred to the target domain network using a frozen transfer technique, automatically finding the optimal network parameters. Furthermore, to enhance the data quality, a data twin composed of the SWSLSTM twin and 1D-ACGAN twin is proposed to capture the dependency relationships of target domain data features. Finally, the migrated network adaptively adjusts the generated samples of the data twin based on the designed indicators, achieving high-precision cross-domain application of the multitwin diagnostic model. The proposed algorithm has been validated utilising data from triple-piston pumps and openly accessible bearing datasets, achieving diagnostic accuracies of 91.6% and 95.28%, respectively. These results substantiate the efficacy of the algorithm, demonstrating its superior generalisation capability and robustness in comparison to extant classical methodologies.

Deep Learning-Based Acoustic Emission Signal Filtration Model in Reinforced Concrete

AbstractAcoustic emission is a nondestructive testing (NDT) technique, widely used to monitor the condition of structures for safety reasons especially in real time. The method utilizes the electrical signals generated by the elastic waves in a material under load to detect and locate damage in structures. However, identifying the sources of AE signals in concrete or composite materials can be challenging due to the anisotropic properties of materials and interpreting a large amount of AE data, leading to data misinterpretation and inaccurate detection of damage. Hence, the need for filtering out noise-induced signals from recorded data and emphasizing the actual AE source is crucial for monitoring and source localization of damage in real time. This study proposed a one-dimensional convolutional neural network (1D-CNN) deep learning approach to filter around 22,000 AE data in a reinforced concrete (RC) beam. The model utilizes significant AE parameters identified through neighborhood component analysis (NCA) to classify true AE signals from noise-induced signals. By using the optimized network parameters, a high classification accuracy of 97% and 96.29% was achieved during the training and testing phases, respectively. To check the reliability of the proposed AE filtering model in the real world, it was evaluated and verified using source location AE activities collected during a four-point bending test on a shear-deficient beam. The outcomes suggest that the proposed AE filtration model has the potential to accurately classify AE signals with an accuracy of 92.8% and proved that the filtration model provides accurate and valuable insight into source location determination.

Learning to solve Bayesian inverse problems: An amortized variational inference approach using Gaussian and Flow guides

Inverse problems, i.e., estimating parameters of physical models from experimental data, are ubiquitous in science and engineering. The Bayesian formulation is the gold standard because it alleviates ill-posedness issues and quantifies epistemic uncertainty. Since analytical posteriors are not typically available, one resorts to Markov chain Monte Carlo sampling or approximate variational inference. However, inference needs to be rerun from scratch for each new set of data. This drawback limits the applicability of the Bayesian formulation to real-time settings, e.g., health monitoring of engineered systems, and medical diagnosis. The objective of this paper is to develop a methodology that enables real-time inference by learning the Bayesian inverse map, i.e., the map from data to posteriors. Our approach is as follows. We parameterize the posterior distribution as a function of data. This work outlines two distinct approaches to do this. The first method involves parameterizing the posterior using an amortized full-rank Gaussian guide, implemented through neural networks. The second method utilizes a Conditional Normalizing Flow guide, employing conditional invertible neural networks for cases where the target posterior is arbitrarily complex. In both approaches, we learn the network parameters by amortized variational inference which involves maximizing the expectation of evidence lower bound over all possible datasets compatible with the model. We demonstrate our approach by solving a set of benchmark problems from science and engineering. Our results show that the posterior estimates of our approach are in agreement with the corresponding ground truth obtained by Markov chain Monte Carlo. Once trained, our approach provides the posterior distribution for a given observation just at the cost of a forward pass of the neural network.

SURVEY ON WBC CLASSIFICATION AND DETECTION USING YOLO

Our heart pumps red blood constantly, carrying waste products, hormones, nutrition, oxygen, and other essential health markers. Blood cells in this fluid are made up of many components, including platelets, RBCs, and WBCs. Neutrophils, Lymphocytes, Monocytes, Eosinophils, and Basophils are further subdivided into WBCs. To precisely identify and treat numerous kinds of blood-related illnesses, the detection and analysis of these blood cells is necessary for advanced diagnostics in medicine. Convolutional neural networks (CNNs), for example, are highly relied upon in modern medical diagnostics for the examination of blood cells. The precision and potency of blood cell identification procedures have been greatly improved by CNNs and Additional approaches to deep learning like RNNs and YOLO. The persistence of this research is to list challenges in White Blood cell detection and classification problem. Deep learning-based blood cell identification and classification approaches based on Single Stage approaches. Single Stage Approach involve You only look Once (YOLO). It explores the various architectures for feature selection and classification starting from YoloV1 toV5 and it’s various modifications. In this paper would be looking at things like network parameters, training procedure, and transfer learning application for each variation. It also assesses model complexity, accuracy, performance, generalization, computational efficiency, and transfer learning capabilities, in addition to variables like data amount and source.

Remote Sensing Image Classification Based on Neural Networks Designed Using an Efficient Neural Architecture Search Methodology

Successful applications of machine learning for the analysis of remote sensing images remain limited by the difficulty of designing neural networks manually. However, while the development of neural architecture search offers the unique potential for discovering new and more effective network architectures, existing neural architecture search algorithms are computationally intensive methods requiring a large amount of data and computational resources and are therefore challenging to apply for developing optimal neural network architectures for remote sensing image classification. Our proposed method uses a differentiable neural architecture search approach for remote sensing image classification. We utilize a binary gate strategy for partial channel connections to reduce the sizes of the network parameters, creating a sparse connection pattern that lowers memory consumption and NAS computational costs. Experimental results indicate that our method achieves a 15.1% increase in validation accuracy during the search phase compared to DDSAS, although slightly lower (by 4.5%) than DARTS. However, we reduced the search time by 88% and network parameter size by 84% compared to DARTS. In the architecture evaluation phase, our method demonstrates a 2.79% improvement in validation accuracy over a manually configured CNN network.

Ensemble Smoother with Fully Convolutional VAE for seismic facies inversion

Seismic facies inversion is an important process in the oil and gas industry to estimate subsurface geological facies or rock types based on seismic data. Recently, the Ensemble Smoother with Multi Data Assimilation (ES-MDA) has shown great success in solving complex inverse problems by generating an ensemble of solutions of the model variables for uncertainty quantification. However, there are significant challenges related to the computational cost, Gaussian assumptions, and its application for categorical model variables. Recent developments have proposed to integrate ES-MDA with Variational Autoencoder (VAE) to reparametrize facies into a continuous feature space and reduce computational costs. Despite promising results, it has introduced new limitations concerning network convergence, spatial correlation loss, and manipulation of the latent space. In this paper, a Fully Convolutional Variational Autoencoder (FCVAE) was proposed to outperform the VAE in terms of preserving spatial information, requiring fewer network parameters, presenting higher dimension reduction and yielding better facies estimation. The objective of the application example is to draw sound conclusions based on a 2D synthetic application, where the results of both methods can be compared with reference facies for quality evaluation.