Neural Network Optimization Algorithm Research Articles

Fault Diagnosis of Rotating Machinery Bearings Based on Improved DCNN and WOA-DELM

A bearing is a critical component in the transmission of rotating machinery. However, due to prolonged exposure to heavy loads and high-speed environments, rolling bearings are highly susceptible to faults, Hence, it is crucial to enhance bearing fault diagnosis to ensure safe and reliable operation of rotating machinery. In order to achieve this, a rotating machinery fault diagnosis method based on a deep convolutional neural network (DCNN) and Whale Optimization Algorithm (WOA) optimized Deep Extreme Learning Machine (DELM) is proposed in this paper. DCNN is a combination of the Efficient Channel Attention Net (ECA-Net) and Bi-directional Long Short-Term Memory (BiLSTM). In this method, firstly, a DCNN classification network is constructed. The ECA-Net and BiLSTM are brought into the deep convolutional neural network to extract critical features. Next, the WOA is used to optimize the weight of the initial input layer of DELM to build the WOA-DELM classifier model. Finally, the features extracted by the Improved DCNN (IDCNN) are sent to the WOA-DELM model for bearing fault diagnosis. The diagnostic capability of the proposed IDCNN-WOA-DELM method was evaluated through multiple-condition fault diagnosis experiments using the CWRU-bearing dataset with various settings, and comparative tests against other methods were conducted as well. The results indicate that the proposed method demonstrates good diagnostic performance.

Can Chinese cities reach their carbon peaks on time? Scenario analysis based on machine learning and LMDI decomposition

As cities are critical actors in mitigating climate change and achieving the “3060″ target, multi-scenario studies on urban carbon emissions can provide a scientific basis for formulating urban carbon peaking action plans. To remedy the problems of missing regional statistics, inconsistent caliber, and lack of city-scale studies in carbon emission research, this paper uses the sparrow optimization neural network algorithm to fit carbon emission data with nighttime stable light for training. Carbon emission data were obtained for 281 cities in China during 2000–2020. The rates of change of influencing factors are set based on shared socioeconomic pathways (SSPs) and representative concentration pathways (RCPs) for different periods and different scenarios. The carbon emission and carbon peaking evolution paths of service, industrial and comprehensive cities from 2021 to 2060 are dynamically simulated. The results show that (1) service cities are significantly higher than industrial and comprehensive cities in population, GDP, secondary industry output, and energy consumption. (2) The economic development effect, as the primary driver of carbon emission growth, increases and then decreases in all five categories of cities, with 2010 as the inflection point. Industrial structure improvement has an increasingly strong offsetting effect on carbon emissions and is one of the critical directions for future carbon emission reduction. (3) Service cities such as Beijing and Shanghai are already at the completion stage of urban transformation and are more likely to reach the carbon peak on their own than other types of cities. In the low carbon following scenario, comprehensive cities such as Kaifeng, Rizhao, and Jilin can achieve their carbon peaking targets efficiently. The findings of this paper can provide valid theoretical support for carbon peaking action programs in China and other countries.

A comprehensive investigation on the influences of optimal CNC wood machining variables on surface quality and process time using GMDH neural network and bees optimization algorithm

The present research work is concerned with the effects of machining parameters on the quality and economical aspects of wooden product. Process variables include depth of cut, feed rate, spindle speed, and step over in three levels and the purpose of the current study is to minimize the surface roughness and the process time, simultaneously. Each variable has three levels and as a result, the number of experimental tests in full factorial mode was considered to 81. Group method of data handling (GMDH) has been employed to train the neural network and extract the relationships governing the machining process. Optimization of the machining parameters has been done using the bees algorithm (BA). The outcomes obtained from GMDH-BA hybrid technique demonstrated that the combined objective function of the surface roughness and process time has been improved about 23% in comparison with the best experiment in all practical tests. The value of surface roughness and the process time in the optimized experimental test were reduced to 4.36 µm and 93 s, respectively.

Synergistic effect of grinding time and submicron (nano) bubbles on the zeta potential state of spent lithium-ion batteries: A gene expression programming approach

The present work aims at the investigation and advanced simulation of the synergistic effect of grinding conditions and sub-micron (nano) bubbles (NBs) on the zeta potential mechanism of spent lithium-ion batteries (LIBs). For this purpose, the variation of the zeta potential of electrode active materials under different conditions was measured based on an effective Historical Data (HD) experimental design. The impact of operating variables including grinding time (0–25 min), pH (4.5–11.5), collector type (collectorless, n-dodecane, kerosene, and diesel oil), and NBs (absence and presence) were assessed through one-way analysis of variance (ANOVA). The process was then simulated using a genetic algorithm (GA) as an optimization algorithm of the artificial neural network (ANN). The statistical results indicated that the process was significantly influenced by pH, collector, and NBs (pvalue < 0.05) through a nonlinear trend. Although the individual effect of grinding time was not significant, a complex nonlinear interaction between grinding time and pH was observed. The effect of collectors followed the order of diesel oil > n-dodecane > kerosene; however, the effect of collectors was inverted at a pH of 10. Finally, the intelligent simulation results revealed that the process could be modeled using a genetic algorithm with a determination coefficient and error of 94.44 % and 3.28 %, respectively.

Stochastic trading of storage systems in short term electricity markets considering intraday demand response market

In recent years, the renewable energy sources are used in the power system for reducing environmental pollution. One of these renewable energy sources is wind farm that is used widely. However, they increase the uncertainty of the power system. To solve this problem, plenty researches have proposed several methods that have some disadvantages. This paper tries to decline the effects of uncertainty of wind turbines by using of the energy storage systems and optimal demand response programs. A stochastic planning of energy storage systems and wind generation are presented in the short term electricity markets considering the trading day-head and intraday market of demand response. In order to formulate an accurate model of various components of system, all needed constraints are considered. In this problem, wind turbines, energy storage systems, and demand response aggregator work coordinately that is called virtual power plant. For better comparison, two states consist of coordinated and uncoordinated working of all components are considered and assessed. The various scenarios for electricity market price, adjustment market price, imbalance energy price, and the generated power of wind turbine are considered. These scenarios are generated with hybrid neural network and coati optimization algorithm. Also, the iterative fast progressive Kantorovich method is used for scenario reduction. The results show the better response of the proposed method in optimal operation of virtual power plant.

Prediction of CO2 emissions in China by generalized regression neural network optimized with fruit fly optimization algorithm.

As global warming becomes more prominent, the need to reduce carbon emissions to achieve China's carbon peak target is increasing. It is imperative to seek effective methods to predict carbon emissions and propose targeted emission reduction measures. In this paper, a comprehensive model integrating grey relational analysis (GRA), generalized regression neural network (GRNN) and fruit fly optimization algorithm (FOA) is constructed with carbon emission prediction as the research objective. Firstly, GRA is used for feature selection to find out the factors that have a strong influence on carbon emissions. Secondly, the parameter of GRNN is optimized using FOA algorithm to improve the prediction accuracy. The results show that (1) fossil energy consumption, population, urbanization rate and GDP are important factors affecting carbon emissions; (2) FOA-GRNN outperforms GRNN and back propagation neural network (BPNN), verifying the effectiveness of FOA-GRNN model for CO2 emission prediction. Finally, by analyzing the key influencing factors and combining scenario analysis with forecasting algorithms, the carbon emission trends in China for 2020-2035 are forecasted. The results can provide guidance for policy makers to set reasonable carbon emission reduction targets and adopt corresponding energy saving and emission reduction measures.

SGDAT: An optimization method for binary neural networks

Stochastic gradient descent (SGD), one of the most popular neural network optimization algorithms, has a solid theoretical foundation as well as good generalization performance. However vanilla SGD performs catastrophically in Binary Neural Networks (BNNs). Many studies have identified this phenomenon without explaining its causes in depth. In this paper, we try to experimentally understand the possible reasons for this and significantly improve the performance of vanilla SGD in BNNs training by adjusting the training strategy to be comparable to Adam. We subsequently propose a new optimization method for training deep neural networks (DNNs) with binary weights. In the proposed SGD with Adaptive Threshold, referred to as SGDAT, we suppress the frequency of weights flipping by thresholds and adjust the threshold of each parameter according to the number of flipping to further reduce the network noise, stabilize the network training, and improve the network generalization ability. Also, we present a complete ablation study of the hyperparameters space, as well as experimentally analyze the impact of using adaptive thresholds. Furthermore, we conduct image classification experiments over the CIFAR10, CIFAR100 and TinyImageNet datasets using BinaryNet and ResNet-18 network structure. The experiments show that SGDAT outperforms other binary optimizers. Code is available at: https://github.com/gushan/SGDAT.

Capacity Consistency Prediction and Process Parameter Optimization of Lithium‐Ion Battery based on Neural Network and Particle Swarm Optimization Algorithm

AbstractThe grading capacity of lithium‐ion battery is an important basis for evaluating battery quality. Aiming at the difficulty of determining the critical control factors and the threshold of the control parameters in the lithium‐ion battery manufacturing process, a capacity prediction and process parameter optimization model of lithium‐ion battery is proposed by combining the back propagation (BP) and particle swarm optimization (PSO) algorithms. First, the BP method is applied to establish the nonlinear mapping relationship between process data and grading capacity, which is regarded as the capacity consistency prediction model. Second, using the prediction model as fitness function and combining with PSO algorithm, the optimization model of process parameters is established. Finally, under the given initial process parameters from the lithium‐ion battery pilot line, it is carried out to obtain the best process parameter formula. The results show that the BP method has an accurate capacity consistency prediction effect. Combined with PSO algorithm, the optimized process parameters are obtained, which significantly improves the capacity consistency of lithium‐ion batteries. The results serve as an engineering application method to guide the selection and confirmation of process parameters at the battery design stage.

Modeling and process parameter optimization of laser cutting based on artificial neural network and intelligent optimization algorithm

Modeling and process parameter optimization of laser cutting based on artificial neural network and intelligent optimization algorithm

Colon Disease Diagnosis with Convolutional Neural Network and Grasshopper Optimization Algorithm.

This paper presents a robust colon cancer diagnosis method based on the feature selection method. The proposed method for colon disease diagnosis can be divided into three steps. In the first step, the images' features were extracted based on the convolutional neural network. Squeezenet, Resnet-50, AlexNet, and GoogleNet were used for the convolutional neural network. The extracted features are huge, and the number of features cannot be appropriate for training the system. For this reason, the metaheuristic method is used in the second step to reduce the number of features. This research uses the grasshopper optimization algorithm to select the best features from the feature data. Finally, using machine learning methods, colon disease diagnosis was found to be accurate and successful. Two classification methods are applied for the evaluation of the proposed method. These methods include the decision tree and the support vector machine. The sensitivity, specificity, accuracy, and F1Score have been used to evaluate the proposed method. For Squeezenet based on the support vector machine, we obtained results of 99.34%, 99.41%, 99.12%, 98.91% and 98.94% for sensitivity, specificity, accuracy, precision, and F1Score, respectively. In the end, we compared the suggested recognition method's performance to the performances of other methods, including 9-layer CNN, random forest, 7-layer CNN, and DropBlock. We demonstrated that our solution outperformed the others.

The property palette: A rapid printing of performance-tunable blended polymers guided by artificial neural network

Existing commercial material extrusion (ME) polymer filaments frequently fail to meet the requirements of actual workpieces and often need improvement through modification. However, traditional modification methods are time-consuming and expensive. Here, we develop a 3D printer capable of blending three polymers and programming the component ratios, which can speed up such a modification process. Additionally, workpieces with varying properties can be quickly produced (just like mixing colors according to the three primary colors) with the guidance of machine learning algorithms. As a demonstration, we used polylactic acid (PLA), thermoplastic polyurethane (TPU), and polyethylene terephthalate glycol (PETG) three commercial filaments as base materials to printout different objects. The tensile strength, tensile modulus, flexural strength, flexural modulus, and hardness of 67 groups of different PLA/TPU/PETG blend components were tested. Based on these data, artificial neural network (ANN) optimization algorithms were used to discover the relationship between the components and properties. The results show that using ANN to predict the mechanical properties of blended polymers can effectively accelerate the development of materials. Materials with tunable properties and complex structures can be manufactured from simple raw materials through 3D printing quickly.

Magnetic Field Feature Fusion of Underwater Vehicle Model Based on Principal Component Analysis

Aiming at the model of underwater vehicle, a magnetic field feature fusion method based on principal component analysis (PCA) is proposed, focusing on: Non-stationary, nonlinear and non-Gaussian magnetic field time domain of underwater vehicle model based on genetic algorithm optimization neural network Feature extraction technology, underwater vehicle model fusion technology based on principal component analysis of magnetic field time-domain features, constructs effective PCA fusion features with low dimension, and finally uses the model sample recording data to conduct experiments on the main methods of this paper Verification and experimental simulation results show that PCA fusion features can reduce the feature dimension on the basis of maintaining a sufficient amount of effective information, and improve the target recognition efficiency. The results confirm the effectiveness of the related algorithms.

Experimental Verification of the Neural Network Optimization Algorithm for Identifying Frequency-Dependent Constitutive Parameters of Viscoelastic Materials

Experimental Verification of the Neural Network Optimization Algorithm for Identifying Frequency-Dependent Constitutive Parameters of Viscoelastic Materials

Neural network optimization algorithm based non-singular fast terminal sliding-mode control for an uncertain autonomous ground vehicle subjected to disturbances

As computer computing capabilities increase, optimization algorithms are becoming more useful for solving engineering problems. Up to now, several metaheuristic algorithms have been exploited in control engineering. However, this effort remains weak in addressing the autonomous ground vehicles (AGVs) trajectory tracking problem. This research presents a novel optimal approach merging the robust non-singular fast terminal sliding-mode control method (NFTSMC) and the neural network optimization algorithm (NNA) for automatic lane changing. First, a reference double lane-change path (DLC) is designed, and the robust non-singular fast terminal sliding-mode steering controller is developed, according to Lyapunov stability theory, to suppress the lateral deviation and ensure the yaw stability. Then, the control strategy is optimized by the NNA algorithm to adjust the steering controller optimally while avoiding local optimums. A comparison, under the same conditions, with the particle swarm optimization algorithm (PSO) revealed the superiority of the control law resulting from the NNA-based optimization. Furthermore, the proposed approach shows its excellent tracking performance versus the integrated backstepping sliding-mode controller (IBSMC) and the adaptive sliding-mode control (ASMC) under severe conditions typical of real-world lane changes.

Https://sifisheriessciences.com/journal/index.php/journal/article/view/1294

Wireless experts worldwide have become interested in using Autoencoders (AEs) for modelling communication systems as an end-to-end reconstruction task. This approach optimizes both the transmitter and receiver components simultaneously, offering flexibility and convenience for representing complex channel models. Traditional communication systems rely on conventional models and assumptions that limit their utilization of limited frequency resources and hinder their ability to adapt to new wireless applications. However, with the rise of Artificial Intelligence, new wireless systems are capable of learning from wireless spectrum data and optimizing their performance. In this paper, the use of deep learning with autoencoders is explored to create an end-to-end communication system that replaces traditional transmitter and receiver activities. The autoencoder architecture effectively addresses channel impairments and enhances overall performance. Simulation results indicate that autoencoders surpass conventional communication systems in terms of Block Error Rate performance, even when facing impairments in the autoencoder's channel layer and using different neural network optimization algorithms.

Projective Fisher Information for Natural Gradient Descent

Improvements in neural network optimization algorithms have enabled shorter training times and the ability to reach state-of-the-art performance on various machine learning tasks. Fisher information based natural gradient descent is one such second-order method that improves the convergence speed and the final performance metric achieved for many machine learning algorithms. Fisher information matrices are also helpful to analyze the properties and expected behavior of neural networks. However, natural gradient descent is a high complexity method due to the need to maintain and invert covariance matrices. This is especially the case with modern deep neural networks, which have a very high number of parameters, and for which the problem often becomes computationally unfeasible. We suggest using the Fisher information for analysis of parameter space of fully connected and convolutional neural networks without calculating the matrix itself. We also propose a lower complexity natural gradient descent algorithm based on the projection of Kronecker factors of Fisher information combined with recursive calculation of inverses, which is computationally less complex and more stable. We finally share analysis and results showing that all these optimizations do not impact the accuracy while considerably lowering the optimization process’s complexity. These improvements should enable applying natural gradient descent methods for optimization to neural networks with a larger number of parameters, than possible previously.

Prediction of Land Use and Landcover Changes in Tiruppur Tamilnadu Using Hybrid Convolutional Neural Network

&lt;p&gt;Land use and landcover change (LU/LC) are important in global change studies because they can transform the local and global environment by developing the biochemical, biochemical and biogeographic properties of the Earth's structure. This paper is intended to develop Hybrid Convolutional Neural Network (HCNN) for land use and land cover changes prediction in Tiruppur Tamilnadu. Initially, the databases are collected from the open-source system. And the image dataset has been pre-processed using the image augmentation technique. Through which the image has been resized and processed for training it with the proposed mode. The resized images are sent to the HCNN for prediction of land cover and land use changes in Tiruppur Tamilnadu. The proposed classifier is a combination of Convolutional Neural Network (CNN) and Remora Optimization Algorithm (ROA). In the CNN, the ROA is utilized to select the hyper parameters to enable efficient prediction in land use and land cover changes. The proposed classifier is implemented in MATLAB and performances is evaluated by performance metrices such as accuracy, precision, recall, sensitivity, F_Measure and Kappa. The proposed methodology is compared with the conventional techniques such as CNN, Markov chain model and Recurrent Neural Network (RNN) respectively.&lt;/p&gt;&#x0D;

Prediction of tool wear in milling process based on BP neural network optimized by firefly algorithm

Long-time discontinuous contact is easy to cause tool wear during milling. To decrease the impact of severe wear on workpiece quality and processing efficiency, cutting tools should be replaced timely. Therefore, tool wear prediction is an important aspect in improving process efficiency, ensuring machining precision and realizing intelligent manufacturing. To boost the precision of online prediction of tool wear, this paper suggests a novel approach to monitor tool wear by optimizing backpropagation (BP) neural network via firefly algorithm (FA). Specifically, the progressive semi-soft threshold function is applied to the process of cutting force signal noise reduction, which reduces redundant signals and noise interference in the signal. Time-domain analysis, frequency-domain analysis, and wavelet packet decomposition are utilized, cutting force features are extracted, and Pearson correlation coefficient is used to sift out signal features that are highly connected to tool wear. The FA is used to improve the BP neural network's weights and thresholds. Through learning nonlinear mapping, relationship between tool flank wear and signal features is realized. A prediction model of tool wear of FA-BP is constructed. Milling experiments validate the prediction model in the milling process. The experimental outcomes confirm the precision and reliability of the method. In comparison to BP neural network, genetic algorithm optimized BP neural network and particle swarm optimization algorithm optimized BP neural network prediction method, it has a greater prediction accuracy and a stronger training impact, and has superior performance. The research results can give a theoretical foundation and technological assistance for predicting tool wear, which is crucial for improving workpiece quality, processing efficiency, and promoting intelligent development.

CFD Modelling and Numerical Simulation of the Windage Characteristics of a High-Speed Gearbox Based on Negative Pressure Regulation

Windage power loss plays a leading role in the total power loss of high-speed gears, which seriously affects the transmission efficiency of gear systems and leads to high energy consumption. This paper proposes a negative pressure regulation method to reduce windage power loss. Based on the computational fluid dynamics theory, the flow field distribution and windage power loss in the gearbox under different negative pressure conditions are studied, and the effect of the negative pressure environment and speed on the windage power loss is obtained. In order to further save calculation costs, an optimization algorithm of the BP neural network based on a genetic algorithm is proposed to effectively predict the windage power loss. The results show that the high-speed airflow near the tooth’s surface will produce a large pressure moment, which is the main cause of wind resistance loss. The windage power loss increases with the increase in the negative pressure or speed of the gearbox, but the effect of speed is more obvious. The prediction results of the optimization algorithm are in good agreement with the finite element simulation data and the open literature, which can predict the best parameters for reducing windage power loss.

Application Model for Innovative Sports Practice Teaching in Colleges Using Internet of Things and Artificial Intelligence

The Internet of Things (IoT) and artificial intelligence (AI) have promoted teaching reform while improving people’s lives. Under the new teaching environment, the position of physical education (PE) teaching in the teaching work has become increasingly prominent. At present, there are some problems in the PE teaching mode of most colleges and universities, such as poor teaching environment, unstable teaching data, and lack of technical support for the teaching system. This also leads to the low quality of PE teaching and unsatisfactory teaching results. In this paper, IoT and AI are combined to study the application mode of innovative practical teaching in college PE. This paper first constructs a physical education teaching system based on the Internet of Things, then summarizes the necessity of artificial intelligence technology participating in the reform of physical education classroom teaching, and gives a specific teaching application model. Finally, based on the golden sine algorithm-optimization neural network, the application model of college physical education in this paper is evaluated. Through experiments and investigations, the new teaching mode improves the teaching efficiency by 14.7%, improves the teaching quality, and provides reference for the next development of IoT and AI in teaching.