Optimal Network Parameters Research Articles

A two-dimensional magnetotelluric deep learning inversion approach based on improved Dense Convolutional Network

Magnetotelluric (MT) inversion is an important means of MT data interpretation. The use of deep learning technology for MT inversion has attracted much attention because it is not limited to the initial model, avoids falling into local optimal solutions, and has the strong ability to process large amounts of data. However, obtaining highly reliable deep learning inversion results remains a challenge. In this paper, we have proposed a two-dimensional (2-D) MT inversion method based on the improved Dense Convolutional Network (DenseNet), with the aim of improving the reliability of the 2-D deep learning MT inversion results. First, the MARE2DEM is used to compute the 2-D MT forward responses when establishing the sample set. Then, an improved DenseNet is proposed by incorporating depthwise separable convolution in lieu of standard convolution within dense connection blocks, and embedding the attention mechanism. Depthwise separable convolution splits the standard convolution operation into depthwise and pointwise convolution, effectively capturing spatial features of input data and correlations between channels. Meanwhile, attention mechanism allows the network to assign varying degrees of importance (or attention) to different elements in a sequence of data, thus enhancing its ability of key feature extraction. This design not only retains the inherent feature reuse and alleviates gradient vanishing of DenseNet but also further enhances network performance. The optimized network parameters for the improved DenseNet are obtained by training on the training set, while the validation set is used to adjust hyperparameters and evaluate model performance. Finally, the proposed 2-D deep learning approach is verified by using both synthetic and field data. Experimental results with synthetic data show that the reliability of inversion results obtained by using the proposed algorithm is improved, and the inversion results obtained by using both TE- and TM-mode data is more accurate than those obtained by using the single mode data. The inversion results of field data show that the proposed 2-D MT deep learning inversion approach can effectively detect the subsurface resistivity structure and has a good application prospect.

Relative pose estimation from panoramic images using a hybrid neural network architecture

Camera-based relative pose estimation (RPE) localizes a mobile robot given a view at the current position and an image at a reference location. Matching the landmarks between views is critical to localization quality. Common challenges are appearance changes, for example due to differing illumination. Indirect RPE methods extract high-level features that provide invariance against appearance changes but neglect the remaining image data. This can lead to poor pose estimates in scenes with little detail. Direct RPE methods mitigate this issue by operating on the pixel level with only moderate preprocessing, but invariances have to be achieved by different means. We propose to attain illumination invariance for the direct RPE algorithm MinWarping by integrating it with a convolutional neural network for image preprocessing, creating a hybrid architecture. We optimize network parameters using a metric on RPE quality, backpropagating through MinWarping and the network. We focus on planar movement, panoramic images, and indoor scenes with varying illumination conditions; a novel dataset for this setup is recorded and used for analysis. Our method compares favourably against the previous best preprocessing method for MinWarping, edge filtering, and against a modern deep-learning-based indirect RPE pipeline. Analysis of the trained hybrid architecture indicates that neglecting landmarks in a direct RPE framework can improve estimation quality in scenes with occlusion and few details.

GAO–FCNN–Enabled Beamforming of the RIS–Assisted Intelligent Communication System

The joint beamforming optimization from the perspective of the bit error rate (BER) in a reconfigurable intelligent surface (RIS)–assisted intelligent communication system is studied in this paper. A genetic algorithm (GA) is investigated to address the bottleneck of the system performance based on the dynamic adaptability theory. However, the bottleneck is caused by the interaction between the active and passive beamforming. To tackle the constraints of conventional optimization approaches, the hybrid scheme is proposed to combine the GA optimization (GAO) and fully connected neural network (FCNN) strategy. Specifically, the intelligent collaborative tuning of system parameters is achieved using this proposed technique. Simulation findings indicate that the hybrid scheme not only simplifies the calculation process to obtain the optimal network parameters, but also effectively optimizes the system structure by dynamically adjusting the RIS reflection configuration. Based on this, the signal transmission quality is improved, interference is reduced, and the stable and efficient operation of the RIS–assisted intelligent communication system is ensured in the complex wireless transmission scenario.

State of Health Estimation of Lithium-Ion Battery Using Multi-Health Features Based on Savitzky–Golay Filter and Fitness-Distance Balance- and Lévy Roulette-Enhanced Coyote Optimization Algorithm-Optimized Long Short-Term Memory

Accurate and reliable state of health (SOH) estimation is extremely crucial for the safe and stable operation of lithium-ion batteries (LIBs). In this paper, a method based on Lévy roulette- and fitness-distance balance-enhanced coyote optimization algorithm-optimized long short-term memory (LRFDBCOA-LSTM) is employed for SOH estimation of LIB. Firstly, six health features are extracted from battery charging and discharging data, and Savitzky–Golay is used to filter the feature data to improve correlation between feature and SOH. Then, Lévy roulette and fitness-distance balance (FDB) strategies are used to improve the coyote optimization algorithm (COA), i.e., LRFDBCOA. Meanwhile, the improved algorithm is used to optimize the internal parameters of long short-term memory (LSTM) neural network. Finally, the effectiveness of the proposed model is comprehensively validated using five evaluation indicators based on battery data obtained under three different testing conditions. The experimental results manifest that after algorithm improvement and network parameter optimization, the performance of the model is significantly improved. In addition, the method has high estimation accuracy, strong generalization, and strong robustness for SOH estimation with a maximum R2 of 0.9896 and minimum R2 of no less than 0.9711.

Optimization of District Heating Network Parameters in Steady-State Operation

Optimization of District Heating Network Parameters in Steady-State Operation

Soft sensor modeling method and application based on TSECIT2FNN-LSTM

To address the issue of low accuracy in soft sensor modeling of key variables caused by multi-variable coupling and parameter sensitivity in complex processes, this paper introduces a TSK-type-based self-evolving compensatory interval type-2 fuzzy Long short-term memory (LSTM) neural network (TSECIT2FNN-LSTM) soft sensor model. The proposed TSECIT2FNN-LSTM integrates the LSTM neural network with the interval type-2 fuzzy inference system to address long-term dependencies in sequence data by utilizing the gate mechanism of the LSTM neural network. The TSECIT2FNN-LSTM structure learning algorithm uses the firing strength of the network rule antecedent to decide whether to generate new rules to improve the rationality of the network structure. TSECIT2FNN-LSTM parameter learning utilizes the gradient descent method to optimize network parameters. However, unlike other interval type-2 fuzzy neural network gradient calculation processes, the error term in the LSTM node parameter gradient of TSECIT2FNN-LSTM is propagated backwards in the time dimension. Additionally, the error term is simultaneously transferred to the upper layer network to enhance network prediction accuracy and memory capabilities. The TSECIT2FNN-LSTM soft sensor model is utilized to predict the alcohol concentration in wine and the nitrogen oxide emission in gas turbines. Experimental results demonstrate that the proposed TSECIT2FNN-LSTM soft sensing model achieves higher prediction accuracy compared to other models.

An optimized BERT fine-tuned model using an artificial bee colony algorithm for automatic essay score prediction

Background The Automatic Essay Score (AES) prediction system is essential in education applications. The AES system uses various textural and grammatical features to investigate the exact score value for AES. The derived features are processed by various linear regressions and classifiers that require the learning pattern to improve the overall score. Issues Moreover, the classifiers face catastrophic forgetting problems, which maximizes computation complexity and reduce prediction accuracy. The forgetting problem can be resolved using the freezing mechanism; however, the mechanism can cause prediction errors. Method Therefore, this research proposes an optimized Bi-directional Encoder Representation from Transformation (BERT) by applying the Artificial Bee Colony algorithm (ABC) and Fine-Tuned Model (ABC-BERT-FTM) to solve the forgetting problem, which leads to higher prediction accuracy. Therefore, the ABC algorithm reduces the forgetting problem by selecting optimized network parameters. Results Two AES datasets, ASAP and ETS, were used to evaluate the performance of the optimized BERT of the AES system, and a high accuracy of up to 98.5% was achieved. Thus, based on the result, we can conclude that optimizing the BERT with a suitable meta-heuristic algorithm, such as the ABC algorithm, can resolve the forgetting problem, eventually increasing the AES system’s prediction accuracy.

Artificial Neural Network Parameter Optimization: Improving Meteorological Data Predictions through Machine Learning

This study aims to create an artificial neural network (ANN) based model to predict solar irradiance using open-sourced meteorological data. A neural network that is feed-forward with backpropagation was employed to build the model. A large combination of model parameters including learning algorithms, transfer functions, number of hidden layers, and neurons was used to customize the neural network. The data used in this study is a part of the publicly available dataset containing real outdoor measurements provided by The National Renewable Energy Laboratory (NREL). The proposed model has been validated by measuring prediction errors using normalized mean squared error (NMSE) and prediction accuracies using regression value (R). The lowest value of the NMSE error was obtained with a neural network model based on three hidden layers employing 40, 8, and 5 neurons respectively. The R-value of this model was the highest among all models. The results have shown that the ascending/descending distribution of neurons in hidden layers is an important factor among other parameters.

Digital twin method for intelligent lifting of cable structures under multi-collaborative mode

In order to ensure the high efficiency and quality consistency of cable structure in each construction stage, this study proposes a digital twin method for intelligent lifting of cable structure under multi-collaborative mode. Firstly, a construction and operation methods of the digital twin logic model are formed. In the logical model, a relationship between the transmission and iteration of construction information in the collaborative process is defined. In order to realize the adaptability of working conditions, the coupling relationship between construction step and structural elevation is established. A cable structure lifting analysis model driven by an improved long short-term memory neural network is proposed, which effectively solves the problem of poor convergence of the traditional finite element method. An automatic time window selection method is formed for the neural network. An adaptive moment estimation method is used for neural network parameter optimization, which improves the efficiency and accuracy of the analysis. Taking the lifting process of a cable truss structure test model as an example, a digital twin collaborative construction platform is established to verify the effectiveness and feasibility of the research method. The results show that the digital twin logic model realizes the process collaboration, scene collaboration and data collaboration, and significantly improves the stability of construction quality and the linkage of construction process. Through the improved long short-term memory neural network, the coupling relationship between the construction step and the structural elevation is established, which accurately guides the whole construction process.

A novel method for detecting cardiac arrhythmias in electrocardiogram signals using the ANFIS classifier

In general, ventricular repolarization instability with Premature Activations (PA) plays a significant role in the detection of arrhythmia using an electrocardiogram (ECG) signal. However, clinical detection of this instability is challenging. In this work, a unique technique for analyzing QT Interval (QTI) dynamics unreliability and other abnormalities in the ECG signal are utilized to detect ventricular tachycardia based on the identified anomalies. Further, a feature extraction method based on Chopped Displacement Gaussian (CDGa) is adopted and the features are retrieved from the CDGa signal and its size is reduced using Fisher Discrimination Analysis (FDA). Also, the results of the FDA are fed into the Adaptive Neuro-Fuzzy interference (ANFIS) classifier. The performance of the ANFIS classifier is enhanced in this study by integrating the Grasshopper optimization algorithm for optimizing network parameters. As a result, the values in the signals are predicted using an adaptive Least Mean Square (LMS)-based prediction algorithm, which is utilized to forecast the future status of VT-affected patients. Results indicate that the proposed work has an overall accuracy of 96 % which is higher than any other conventional methods. Also, the various additional performance criteria are tested, and it is demonstrated that the CDGa with adaptive LMS outperforms all previous approaches.

Disentangled variational auto-encoder for multimodal fusion performance analysis in multimodal sentiment analysis

Multimodal Sentiment Analysis (MSA) holds extensive applicability owing to its capacity to analyze and interpret users' emotions, feelings, and perspectives by integrating complementary information from multiple modalities. However, inefficient and unbalanced cross-modal information fusion substantially undermines the accuracy and reliability of MSA models. Consequently, a critical challenge in the field now lies in effectively assessing the information integration capabilities of these models to ensure balanced and equitable processing of multimodal data. In this paper, a Disentanglement-based Variable Auto-Encoder (DVAE) is proposed for systematically assessing fusion performance and investigating the factors that facilitate multimodal fusion. Specifically, a distribution constraint module is presented to decouple the fusion matrices and generate multiple low-dimensional and trustworthy disentangled latent vectors that adhere to the authentic unimodal input distribution. In addition, a combined loss term is modified to effectively balance inductive bias, signal reconstruction, and distribution constraint items to facilitate the optimization of neural network weights and parameters. Utilizing the proposed evaluation method, we can evaluate the fusion performance of multimodal models by contrasting the classification degradation ratio derived from disentangled hidden representations and joint representations. Experiments conducted with eight state-of-the-art multimodal fusion methods on the CMU-MOSEI and CMU-MOSEI benchmark datasets demonstrate that DVAE is capable of effectively evaluating the effects of multimodal fusion. Moreover, the comparative experimental results indicate that the equalizing effect among various advanced mechanisms in multimodal sentiment analysis, as well as the single-peak characteristic of the ground label distribution, both contribute significantly to multimodal data fusion.

A detection method of Auricularia auricula mycelium storage period based on near-infrared spectral characteristics and deep learning model

The storage time of Auricularia auricula mycelium was an important factor influencing its growth activity. The rapid detection of storage time had important theoretical significance and practical value for evaluating, cultivating, and scientifically managing Auricularia auricula germplasm resources. Due to the problem of mapping the relationship between the complex near-infrared (NIR) spectral characteristics of Auricularia auricula mycelium and its storage period, it was difficult to characterize using mathematical models accurately. In this paper, a method for detecting the storage period of Auricularia auricula mycelium based on NIR spectral characteristics and deep learning was proposed. First, five different storage periods of Auricularia auricula mycelium were taken as the research object, and the de-trending (DT) algorithm was used to pre-process the spectral data. Then, the Competitive Adaptive Reweighted Sampling (CARS) algorithm was applied to extract the NIR spectral characteristics of the mycelium. The reduction rate of the spectral variables reached 94.64 %. Further, a one-dimensional convolutional neural network (T-CNN) was constructed, and the SGD-Adam algorithm was used to calculate the optimal network parameters. Finally, a method for detecting the storage period of Auricularia auricula mycelium was implemented, achieving a simulation accuracy of 98.33 %. The experimental results revealed that this method outperformed traditional machine learning models such as partial least squares-discriminant analysis (PLS-DA), back propagation (BP), support vector machine (SVM), and k-nearest neighbors (KNN) by an average of 19.56 % in detection accuracy. Additionally, the detection speed was only 0.07 s. This achievement enabled rapid and accurate detection of the storage time of Auricularia auricula mycelium, providing a dependable foundation for identifying the growth activity of Auricularia auricula mycelium. It provided technical support for the scientific management of the edible fungus mycelium, including quality monitoring, evaluation, and screening.

Thermal displacement prediction of high-speed electric spindles based on BWO-BiLSTM

To accurately, efficiently and stably predict the thermal displacement of the spindle, a prediction model based on the beluga whale algorithm (BWO) optimized bi-directional long and short-term memory neural network (BiLSTM) is introduced in this paper. Firstly, the thermal characterization and simulation analysis of the spindle are carried out, and the temperature and thermal displacement change characteristics of the spindle are obtained. Then the thermal deformation experiment of the spindle is carried out, and the temperature and displacement sensors are set up reasonably according to the temperature and thermal displacement change characteristics of the spindle, and the experimental data are collected and analyzed. The adaptive and globally convergent BWO is selected to optimize network parameters of BiLSTM, and the BWO-BiLSTM prediction model is constructed by learning the nonlinear correlation characteristics between spindle temperature and axial thermal displacement. The constructed BWO-BiLSTM prediction model is compared with other prediction models, and it is found through analysis that the prediction results output from the BWO-BiLSTM model have better accuracy and stability. The results of the study can provide a certain theoretical basis and technical support in predicting the spindle thermal displacement, which can help to promote the precision machining production of electric spindles.

Deep domain-adversarial anomaly detection with robust one-class transfer learning

Establishing an anomaly detection model as quickly as possible with a small number of measurement samples for use with new equipment or working conditions has been a research focus. In recent years, despite the effectiveness of transfer learning techniques in terms of coping with such problems, it is still challenging to map the detection rules of an existing device to a target device to enhance its single-class anomaly detection capabilities. In addition, the robustness requirement of the detection rule transfer process makes the problem further complicated, particularly in cases where limited target device data still have unknown outliers. To achieve good anomaly detection performance in such a complex scenario, we propose a robust deep one-class support vector description-based anomaly detection algorithm built on adversarial transfer learning. First, we designed a new hyperspherical adversarial training mechanism that integrates robust adaptive constraints into an improved domain-adversarial neural network. Additionally, to eliminate outlier interference to the greatest extent possible during the domain-adversarial process, we aimed to compute the probability of a sample being normal based on the utilized anomaly detection module and embedded this probability in a domain discriminator. Finally, an end-to-end optimization strategy was derived to find the optimal network parameters that could separate the anomalous classes as much as possible while promoting the consistency of the normal class data in the source and target domains. To validate the performance of our model, we experimented with three scenarios, i.e., digit image detection (with the MNIST and USPS datasets), object recognition (with the Office-Home dataset), and rolling bearing anomaly detection. The results showed that the proposed algorithm outperformed the state-of-the-art methods in terms of detection accuracy and robustness.

Fault diagnosis of drivetrain systems based on vibration signals and convolutional neural networks

Accurate and fast fault diagnosis can improve the reliability of drivetrain systems. The fault diagnosis model based on convolutional neural networks is tricky because the accuracy and efficiency of diagnosis depend on the neural network parameters, such as layer size and number of layers. In this paper, we developed a data pre-processing technique to format vibrational data to be receptible to a neural network. First one will study the difference in performance by inputting a spectrogram of the vibrational data compared to the raw data itself. Second, the windowizing technique is used and its parameters are explored to train different configurations of convolutional neural networks. Finally, the optimal neural network parameters are obtained to improve the computational efficiency while the training model has the highest accuracy.

Deep learning with local spatiotemporal structure preserving for soft sensor development of complex industrial processes

Data-driven soft sensors have emerged as indispensable tools for predicting quality variables in complex industrial processes because of their cost-effectiveness and ease of maintenance. In particular, soft sensors based on deep learning have been utilized in extensive research and successful applications in recent years. However, traditional deep learning methods capture hierarchical data features by minimizing global fitting errors, neglecting the local structural characteristics implied in the original data. In this paper, we propose a new deep learning method for soft sensor development. Utilizing autoencoders as the foundational architecture of our network, a new semisupervised strategy is adopted for layerwise pretraining optimization. On the one hand, more representative data features are extracted by maintaining the local spatiotemporal structure of the data; on the other hand, layer-by-layer supervised learning is employed to identify the critical features that are aligned with the ultimate task, which aids in obtaining the optimal network parameters and improving the resulting prediction accuracy. Subsequently, a local spatiotemporal structure-preserving stacked semisupervised autoencoder (LSP-SuAE) is established. To evaluate the feasibility and effectiveness of the proposed approach, experiments are carried out in a real industrial process. A soft sensor based on the LSP-SuAE is developed to predict the rate of ethylbenzene conversion during the dehydrogenation of styrene. The experimental results demonstrate that, compared to five other common or similar data-driven modeling methods, LSP-SuAE exhibits higher prediction accuracy and better stability.

Calculation of electrical energy losses in distribution networks for power supply using oil and vacuum circuit breakers

A method for approximate estimation of load losses in distribution networks from 6 to 10 kV, as well as for lower voltage networks, has been proposed. The main advantage of this method is that the root mean square current is calculated only once for a series of calculations. This approach significantly reduces the time required for calculations and improves their accuracy, as it eliminates the need for multiple determinations of the root mean square current for each individual calculation. The proposed approach contributes to a better understanding of the influence of various factors on load losses, which allows for more effective planning and management of distribution networks. This is particularly important in the context of increasing loads on electrical networks and higher demands for their reliability and efficiency. In the calculation of electrical energy losses in distribution networks from 6 to 10 kV, methods that use generalized coefficients of electrical circuits and operating modes in the form of regression equations are widely applied. The most significant and independent factors, such as the input of active energy, total line length, number of line sections, total number of transformers, and their total installed capacity, must be taken into account and are included in the calculation equation. Additionally, the components used and their regular maintenance play an important role in electrical energy losses in electrical networks. The main direction of research is the optimization of distribution network parameters, taking into account the latest technologies and materials, as well as the implementation of automated monitoring and control systems that will allow timely detection and elimination of problems associated with electrical energy losses. The proposed method for estimating load losses in distribution networks is a promising direction for the development of the energy sector, which will contribute to improving the efficiency and reliability of electrical networks, as well as reducing the costs of their operation and maintenance.

Precision Mars Entry Navigation with Atmospheric Density Adaptation via Neural Networks

Spacecraft entering Mars require precise navigation algorithms capable of accurately estimating the vehicle’s position and velocity in dynamic and uncertain atmospheric environments. Discrepancies between the true Martian atmospheric density and the onboard density model can significantly impair the navigation performance. This work introduces a new approach to online filtering for Martian entry using a neural network to estimate atmospheric density and employing a “consider” analysis to account for the uncertainty in the estimate. The network is trained on an exponential atmospheric density model, and its parameters are dynamically adapted in real time to account for any mismatch between the true and estimated densities. The adaptation of the network is formulated as a maximum likelihood problem by leveraging the measurement innovations of the filter to identify optimal network parameters. Within the context of the maximum likelihood approach, incorporating a neural network enables the use of stochastic optimizers known for their efficiency in the machine learning domain. Performance comparisons are conducted against two online adaptive approaches, covariance matching and state augmentation and correction, in various realistic Martian entry navigation scenarios. The results show superior estimation accuracy compared to other approaches and precise alignment of the estimated density with a broad selection of realistic Martian atmospheres sampled from perturbed Mars Global Reference Atmospheric Model data.

Geometrically nonlinear bending analysis of laminated thin plates based on classical laminated plate theory and deep energy method

This paper establishes a geometrically nonlinear bending analysis framework using the deep energy method and the classical laminated plate theory (CLPT) for laminated plates. Inspired by the transfer learning technique, a load applied to a laminated plate can be divided into multiple load steps. The network parameters for the current load step, with the exception of the initial step, are initialized by inheriting values from their preceding steps. Including both von Kármán and Green-Lagrange strains, the plate strains are computed using the automatic differentiation and integrated along the thickness direction per laminate plate based on the constitutive theory. By combining the outputs of neural network, the external potential energy can be obtained, and the optimized network parameters are given by minimizing the total system potential energy of the laminated plate. In order to validate the proposed approach, several numerical examples are calculated, and the present solutions are compared with those given by the literature and the Finite Element Analysis (FEA). The results show that the proposed approach is indeed feasible, can reach high levels of precision under varying loads while offering a simplified calculation strategy.

Real-time intelligent 3D holographic photography for real-world scenarios.

Three-dimensional (3D) display can provide more information than two-dimensional display, and real-time 3D reconstruction of the real-world environment has broad application prospects as a key technology in the field of meta-universe and Internet of Things. 3D holographic display is considered to be an ideal 3D display scheme, thus enhancing the computational speed and reconstruction quality of 3D holograms can offer substantial support for real-time 3D reconstruction. Here, we proposed a real-time 3D holographic photography for real-world scenarios driven by both physical model and artificial intelligence. The 3D information of the real scene was acquired by a depth camera and then divided into 30 layers using the layer-based method. Convolutional neural networks (CNN) were used to build the mapping of intensity and depth maps to computer-generated holograms (CGH). The differentiability of the angular spectrum algorithm was used to realize the self-supervised training of the network, while the composite loss function was employed to optimize network parameters by calculating the loss between reconstructed and target images. The trained network can generate a CGH with a resolution of 1024×1024 in 14.5 ms. The proposed system operates at 22 frames per second and successfully reconstructs 3D video of dynamic scene. The system exhibits significant potential for application in intelligent manufacturing, remote office work, distance education and other fields.