Prediction Results Of Neural Network Research Articles

Research on the Influence of Wheel Tread Defects on Rolling Contact Fatigue

With the rapid development of high-speed trains, local rolling contact fatigue caused by special local defect damage forms has become more prominent. This article takes circular defects as the research object, adopts the Jiang Sehitoglu multi-axis fatigue damage criterion based on the critical plane method, establishes a finite element model of wheel tread circular defects, and studies the influence of wheel tread circular defects on wheel rail rolling contact fatigue. Based on finite element analysis to obtain fatigue damage parameters, an improved PSO-BP neural network was used to establish a neural network prediction model for fatigue damage parameters, and the feasibility of the model was demonstrated. The research results show that the main influencing factors on the rolling contact fatigue life of wheel tread defects are shear stress and shear strain; The fatigue damage parameter is maximum at the edge of the wheel tread defect; As the defect distribution moves from the center to both sides, the contact stress and damage parameters gradually decrease; As the depth of the defect increases to the size of the radius, the contact stress and damage parameters first increase and then decrease; As the diameter of the defect increases, the contact stress and damage parameters increase, but the amplitude change gradually tends to stabilize with the increase of the defect diameter. The neural network prediction results indicate that all predicted samples are within a reasonable range, and this neural network model can provide a reference for predicting fatigue damage of wheel tread.

Research on prediction of slurry migration distance in aggregate stacking based on GA-PSO-BPNN algorithm

After the water inrush accident in coal mine tunnels, early-stage pouring of aggregate forms a high-resistance, low-permeability aggregate stacking, transforming the pipeline flow into percolation. In the later stage, grouting is carried out into the interior of the aggregate stacking, effectively accumulating and solidifying the cement slurry. Among these, whether the slurry can migrate over long distances and fill the voids inside the aggregate stacking is the critical determinant of the success or failure of sealing. To quantitatively analyze the migration distance of slurry inside the aggregate stacking after grouting, a single-hole grouting test platform was established, and an orthogonal experiment was designed with grouting pressure, water cement ratio, and aggregate stacking porosity as influencing factors. Based on 25 sets of experimental measurements, four neural network prediction models suitable for studying the slurry migration distance within the aggregate stacking were constructed separately as back propagation neural network (BPNN), genetic algorithm (GA) combined BPNN, particle swarm optimization (PSO) combined BPNN, and GA-PSO combined BPNN. Evaluation criteria such as Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Mean Square Error (MSE), Root Mean Square Error (RMSE), and the coefficient of determination (R2) were used for comparative analysis of the calculation errors and prediction accuracy of each model. From the perspective of neural network prediction results, the weight value of each influencing factor was analyzed, and the ranking was as follows: grouting pressure > aggregate particle size > water cement ratio, with grouting pressure being the primary controlling factor. The study demonstrates that the GA-PSO-BP model exhibits the best prediction performance, with an average relative error of only 1.59% and an R² of 0.998. This neural network model overcomes issues such as slow learning and getting stuck in tricky spots in BP neural networks. The prediction model shows high accuracy and stability, enabling more effective and accurate prediction of slurry migration distances, making it worthy of dissemination and application. This study can improve safety measures by reducing waste, expediting disaster management efforts, and minimizing environmental hazards associated with mining incidents.

Rapid optimization method of solid rocket motor propellant with structural strength

AbstractTo design a solid rocket motor with a high mass ratio, it is necessary to find out the parameters of the grain shape to meet the structural strength under volume loading fraction and initial burning areas conditions. A rapid optimization method for complex grain of solid rocket motors based on parametric modelling and GA‐BP is proposed. Through the sensitivity analysis of the geometric parameters related to the grain, the sensitive parameters are determined. 30 groups of parameters are determined by the mixed horizontal orthogonal test and the learning samples of the neural network are obtained by FEA. The artificial neural network is established based on the FEA results, and the accuracy of the network is verified by 10 groups of test data sampled by the LHS method. The comparison results show that the prediction error of the maximum Von Mises strain is 5.09 %, the prediction error of the volume loading fraction is 0.16 %, and the prediction error of the initial burning surface is 0.865 %. Based on this, the neural network prediction results are optimized as a function of fitness in GA, and the optimization results show that the maximum strain is reduced by 20.41 % while the constraints are satisfied.

Rapid Optimization Method of Solid Rocket Motor Propellant Under Multiple Load Condition

In order to design a solid rocket motor with a high mass ratio, it is necessary to find out the parameters of the grain shape to meet the structural strength under volume loading fraction condition. A rapid optimization method for a complex grain of solid rocket motors based on parametric modeling and GA-BP is proposed. Based on this, the neural network prediction results are optimized as a function of fitness in GA, and the optimization results show that the maximum Von Mises strain is reduced by 9.8% while the constraints are satisfied.

Energy financial risk early warning model based on Bayesian network

Oil is a global, non-renewable energy source, which plays a pivotal role in the development of the global economy and the strategic reserve system. With the expansion of crude oil futures trading scale, crude oil is no longer a pure energy commodity, but is increasingly controlled by the financial market. Based on Bayesian network, this paper studies and analyzes the financial risks of non renewable energy such as oil. Finally, the prediction models of BP neural network and Bayesian network are given. Neural network is a mathematical model based on neurons, a nonlinear adaptive dynamic system formed by the connection between neurons, and based on these two models, discusses the financial risk early warning method of petroleum in non-renewable energy. The test results of this paper show that the Bayesian network and BP neural network are used for early warning analysis based on the comprehensive oil financial risk data from 1987 to 2015. The prediction results of neural network show that there is 65% correlation between the two. From the fitting results of the prediction results and real data, the fitting effect of Bayesian network is better than that of BP neural network, and the fitting success rate is as high as 80%. The Bayesian network prediction model is analyzed by using the autoregressive movement of the model itself and the time series measurement results of the data itself. The results show that the prediction model is highly reliable. Therefore, it is very necessary to adopt Bayesian network prediction model in the risk early warning of petroleum finance.

A Novel Hybrid Methodology to Study the Risk Management of Prefabricated Building Supply Chains: An Outlook for Sustainability

The management of the prefabricated building supply chain involves the entire process of prefabricated buildings. There are many uncertain factors, and the risk factors in any link will affect the overall operation of the supply chain. In order to achieve the “dual carbon” goal as soon as possible and promote the sustainable development of the building supply chain, it is very important to study the risk management of the assembly building supply chain. The risk management of the prefabricated building supply chain involves risk recognition, risk prediction, risk assessment, and risk response. In this study, on the basis of literature research, the WBS-RBS (Work Breakdown Structure–Risk Breakdown Structure) method comprehensively uses the working link and risk type of the prefabricated building supply chain to establish an indicator system for risk factors in the prefabricated building supply chain. Then, the risk prediction and evaluation model of the neural network of BP (Back Propagation) through Python software is established to predict the risk of prefabricated building supply chains. After verification, it was found that the accuracy of the training set and test set reached 100% and 96.6667%. The results showed that the BP neural network had good effects on the risk forecast of the prefabricated building supply chain, which provided certain risk predictions for the risk prediction of the prefabricated building supply chain. For reference, on the basis of risk prediction, in order to explore the importance of risk factors to the results of BP neural network prediction results, the characteristic importance algorithm of machine learning replacement features further analyzes the risk factors of the prefabricated building supply chain. Finally, based on the prefabricated construction project of enterprise A, risk prediction and evaluation of its supply chain management were carried out, countermeasures for targeted risks were proposed, and we provided new research on the sustainable development of the assembled building supply chain to provide new research ideas.

Research on stamping forming prediction of aluminum alloy sheet based on RBF neural network

In order to accurately predict and reduce the possible defects in the stamping process of an aluminum alloy sheet, the simulation data of the sheet thickness for the 6016 aluminum alloy in the stamping process were obtained by the Hill’48 yield criterion based on finite element ABAQUS/Explicit solver. Taking blank holder force, friction coefficient, stamping speed, and die clearance as input parameters, the radial basis function (RBF) network model for predicting the maximum thinning rate of the stamping aluminum alloy sheet was established. The results show that the RBF network model constructed in this paper has high precision and can reflect the complex relationship between the stamping process parameters and the maximum thinning rate well by comparing the finite element simulation and neural network prediction results. It is of great significance to improve the optimization efficiency of the stamping process of the aluminum alloy sheet and reduce the actual experimental cost.

A comparative analysis of maximum likelihood estimation and artificial neural network modeling to assess electrical component reliability

AbstractThis study focuses on accurately predicting the behavior of new power function distribution using neural network and optimizing it using maximum likelihood estimation. The main motivation of this study is that there is no study in the literature that optimizes and predicts the reliability analysis of lifetime models by combining artificial neural networks and maximum likelihood estimation methods. The numerical findings of the reliability investigations and the values got from maximum likelihood estimation and artificial neural network modeling have been examined and investigated carefully. For the artificial neural network models, the R value was 0.99999 and the deviation ratios were lower than 0.08%. The findings reveal that artificial neural networks are a powerful and useful mathematical tool for analyzing the reliability of lifetime models and numerical study findings via maximum likelihood estimation are completely in accord with artificial neural network prediction results.

Study on stability of cavity in metal–organic chemical vapor deposition calculation based on neural network method

The computational fluid dynamics (CFD) method is widely used to study the process parameters and internal flow states of reactor chambers based on metal–organic chemical vapor deposition (MOCVD) to guide film growth. Currently, several machine learning models have been used in CFD studies, and the prediction accuracy of such models is positively correlated with the amount of data. Thus, two-dimensional (2D) models are used in CFD studies, while three-dimensional (3D) models contain more information and have been used more widely. Herein, neural network (NN) models for target regions based on a 3D MOCVD reactor are proposed and applied to flow-stability studies using the MOCVD reactor chamber. NN models are used to predict the cavity stability curve, and the range of process parameters can be controlled by the characteristics of the curve. NN prediction results have higher accuracy, after the model is established, which considerably reduces the work of CFD numerical simulation and lays a foundation for MOCVD equipment design and process debugging.

Optimal Sprinkler Application Rate of Water–Fertilizer Integration Machines Based on Radial Basis Function Neural Network

The application rate for sprinkler irrigation of water–fertilizer integration machines is an important technical parameter for efficient operation. If the value is too large, the equipment will produce runoff; if it is too small, the equipment will run too long and waste energy. Therefore, it is necessary to provide a feasible scientific and theoretical basis for developing a reasonable application rate. In this study, a mathematical model of soil infiltration for sprinkler irrigation with water and fertilizer integration machines was developed. Soil water accumulation time for different soil’s initial water content, bulk density, sprinkler application rate and soil texture were derived by the finite element method, and these data were used as a training database for the neural network. To make the neural network convenient for predicting the optimal application rate of sprinkler irrigation (the maximum application rate of sprinkler irrigation without runoff) in practice, the time of waterlogging, was multiplied by the optimal application rate of sprinkler irrigation to obtain the total irrigation volume. The optimal application rate of the sprinkler irrigation prediction model of radial basis function (RBF) neural network was constructed with total irrigation water, soil bulk density, initial water content and soil texture as inputs and compared with BP neural network and generalized regression neural network. The highest prediction accuracy of RBF neural network was obtained, and its average relative error was 0.11. To verify the accuracy of the RBF neural network application rate of sprinkler irrigation prediction model in real life, a sprinkler experiment was conducted in the laboratory of Guangzhou University, and the collected soil and lawn of Guangzhou University were used to simulate the actual environment. The results showed that the relative error between the RBF neural network prediction results and the actual values was generally around 10%, while for a total irrigation volume of 58 mm, the optimal application rate of sprinkler irrigation calculated with the model was 42 (mm/h), which can save 70% of irrigation time compared to the case of using the stable infiltration rate of soil as the application rate of sprinkler irrigation without water and fertilizer. Water and fertilizer losses were not observed. This indicates that the model proposed in this study is of practical value in determining the optimum application rate of sprinkler irrigation for water–fertilizer integration machines.

Warning model of new energy vehicle under improving time-to-rollover with neural network

The probability of electric vehicle rollover accident can be effectively reduced by shortening the prediction time interval and improving the prediction accuracy. Based on a multilayer neural network, an improved time-to-rollover method is presented in this paper. Firstly, the force model of vehicle rollover is established and analyzed where the structure and mass of a battery box have an important influence on the occurrence of rollover. Then, the rollover indexes considering hyperparameters are divided into five categories, and the multi-layer neural network is used to simplify the algorithm structure of the time to rollover, and quickly calculate the operating state parameters with a variation step size in real time. Finally, the influence of the hyperparameters on the prediction results of neural network is studied, and higher efficiency is obtained by comparing with traditional methods.

Research on Prediction Model of Rock and Soil Layer Information Based on Adjacent Boreholes

Aiming at the problem that the drilling core cannot be drilled and core is unable to be drilled to find the stratum due to the construction measures or large equipment occupation in the survey site, based on the thickness information of the adjacent hole position of the borehole to be measured, the BP neural network model for stratum prediction was constructed, and the training data sampling method and comprehensive analysis results were put forward, which were based on the training data sampling method and comprehensive analysis results in the front, back, left, right and circle positions of the borehole, combined with the actual investigation Engineering. The influence of soil layer training data collected by different methods on the accuracy of soil layer thickness prediction was analyzed. The results showed that: (1) By comparing the prediction results of 10 groups of actual engineering holes to be measured, it was revealed that the prediction accuracy of soil thickness had great correlation with the orientation selection. (2) In azimuth prediction, each group collects 21 adjacent boreholes as training data samples by continuous cycle of three longitudinal and seven horizontal directions. The prediction results of neural network were stable. (3) A selection strategy was proposed to analyze the preliminary prediction results of neural network. Based on the circular prediction results, the data with the difference within ±0.2m in azimuth prediction were compared and screened. The comprehensive prediction value obtained by test was very close to the true value, and the data error is within 9%, and the prediction effect was good. The research results provide a way of thinking for the prediction of small sample soil sequence, and can provide reference for geotechnical engineering investigation and design application.

Many-objective rapid optimization of reactor shielding design based on NSGA - III

In this paper, a many-objective optimization method for reactor shielding design coupled with NSGA - III and neural network is proposed. By establishing the many-objective optimization model, using Monte Carlo method to calculate samples to train neural network, and the prediction results of neural network are used as the parameters of fitness function for many-objective optimization. The coupling between neural network and NSGA - III is realized, and the Pareto optimal solution of many-objective optimization of reactor shielding design is obtained. The results show that the method of neural network coupling NSGA - III performs well in solving the many-objective optimization problem, and can be applied to the many-objective optimization engineering design of reactor radiation shielding.

Combined Prediction of Photovoltaic Power Based on Sparrow Search Algorithm Optimized Convolution Long and Short-Term Memory Hybrid Neural Network

To address the problem of strong uncertainty in the high proportion of new energy output, an improved convolutional long- and short-term memory (CLSTM) hybrid neural network is proposed for PV power combination prediction. Firstly, considering the large impact of weather changes on PV power output, a fluctuation feature identification model is used to classify historical PV power series samples into slow weather change type and severe weather change type. Secondly, taking into account the multimodal characteristics of PV power output, an improved variational modal decomposition technique is used to adaptively determine the number of modal components, K, and decompose the two types of samples. Regarding the existence of the low-frequency steady state component and the high-frequency fluctuation component of PV power output, the high-frequency component is used to train the long- and short-term memory (LSTM) model and the low-frequency component is used to train the convolutional neural network (CNN) model. The improved sparrow search algorithm (SSA) is used to optimize the parameters of the LSTM and CNN models during the training process. Finally, the predicted component values of each model are superimposed and reconstructed to obtain PV power prediction values. The actual operation data of a PV plant in northern China were used for comparison and validation, and the experiments showed that the accuracy of the prediction results, based on the improved SSA to optimize the parameters of the CLSTM hybrid neural network for predicting PV output, was significantly better than that of the BP, CNN, LSTM single neural network prediction results, and of the prediction accuracy of the unoptimized CLSTM hybrid neural network. At the same time, compared with the above single neural network and unoptimized hybrid prediction model, the proposed method converged faster and was more adaptable to weather changes.

Semantic Analysis of Multimodal Sports Video Based on the Support Vector Machine and Mobile Edge Computing

Even though sports video analysis and research have yielded some results, the competition video itself contains a large number of images, video, audio, text, and other elements. Video semantic analysis has always been a hot and difficult topic in video studies. Multimodal sports video data contain a wealth of data. Second, it has a complex structure with an ambiguous relationship between each video unit. A typical algorithm based on kernel function learning is a support vector machine. It has shown promising results in the prediction of landslide displacement time series. The semantic analysis of multimodal sports video using support vector machines and mobile edge computing is the subject of this study. The theoretical prediction results based on the monitoring data of multimodal sports video semantic analysis show that they are comparable to neural network prediction results using genetic algorithm. The semantic analysis of multimodal sports video based on support vector machine method proposed in this study has a better prediction ability, and its theoretical prediction results are close to the actual monitoring values.

ReF-nets: Physics-informed neural network for Reynolds equation of gas bearing

Based on the deep learning method, the physical information of Reynolds equation is introduced into the neural network, and a deep learning network frame (ReF-nets) is built to predict the flow fields and aerodynamic characteristics of gas bearing. The prediction results of neural network have high enough accuracy comparing with those from finite difference method (FDM). Furthermore, a comprehensive investigation of unsupervised learning, supervised learning and semi-supervised learning in PDE (partial differential equation) solution prediction is carried out, which is the first time these three learning strategies are fully compared in PDE solving field to our best knowledge. The advantages and disadvantages of the three learning methods are illustrated in detail from the perspectives of physics interpretability and prediction accuracy. It is found that the unsupervised learning method has the strongest physics interpretability in predicting the flow field. The supervised learning method has the highest prediction accuracy, but it has nearly no physics interpretability. While semi-supervised learning method can take into account the advantages of both methods, and perform well when there are some known flow field data.

Landslide Displacement Prediction Method Based on GA-Elman Model

The deformation process of landslide displacement has complex nonlinear characteristics. In view of the problems of large error, slow convergence and poor stability of the traditional neural network prediction model, in order to better realize the accurate and effective prediction of landslide displacement, this research proposes a landslide displacement prediction model based on Genetic Algorithm (GA) optimized Elman neural network. This model combines the GA with the Elman neural network to optimize the weights, thresholds and the number of hidden neurons of the Elman neural network. It gives full play to the dynamic memory function of the Elman neural network, overcomes the problems that a single Elman neural network can easily fall into local minimums and the neuron data is difficult to determine, thereby effectively improving the prediction performance of the neural network prediction model. The displacement monitoring data of a slow-varying landslide in the Guizhou karst mountainous area are selected to predict and verify the landslide displacement, and the results are compared with the traditional Elman neural network prediction results. The results show that the prediction results of GA-Elman model are in good agreement with the actual monitoring data of landslide. The average error of the model is low and the prediction accuracy is high, which proves that the GA-Elman model can play a role in the prediction of landslide displacement and can provide reference for the early warning of landslide displacement deformation.

Analysis of Sports Performance Prediction Model Based on GA-BP Neural Network Algorithm.

There are many factors that affect athletes' sports performance in sports competitions. The traditional sports performance prediction method is difficult to obtain more accurate sports performance prediction results and corresponding data analysis in a short time, which is not conducive for coaches to formulate targeted and scientific training sprint plans for athletes' problems. Therefore, based on GA-BP neural network algorithm, this paper constructs a sports performance prediction model and carries out experiments and analysis. The experimental results show that GA-BP neural network algorithm has a faster convergence speed than BP neural network and can achieve the expected error accuracy in a shorter time, which overcomes the problems of the BP neural network. At the same time, different from the previous models, GA-BP neural network algorithm can get the athlete training model according to the relationship between quality training indicators and special sports training results, which can more intuitively show the advantages and disadvantages of athletes. In the final sports performance prediction results, GA-BP neural network prediction results have higher accuracy, better stability, better prediction effect, and higher application value than BP neural network.

Spunlace (Su Jeti ile Bağlama) Teknolojisi ile Üretilen Dokusuz Yüzeylerin Morfolojik Özelliklerinden Bazı Performans Özelliklerinin Yapay Zeka ile Tahminlenmesi

Fiber placement and fiber distribution characteristics of nonwoven surfaces significantly affect the physical, mechanical and permeability properties of the fabric. In the studies from the literature, significant relationships between fiber distribution and porosity with mechanical performance properties have been revealed. In this study, an algorithm developed using image processing techniques for statistical data related to texture features were obtained images from nonwoven surface fabric samples. The texture features obtained were used as input data in the artificial neural network model. Air permeability, machine direction breaking strength, cross direction breaking strenth, and breaking elongation performance characteristics were used as output data. Thus, it is aimed to estimate the air permeability, breaking strength and breaking elongation performances of the fabric samples produced with spunlace (hydroentaglement bonding) technology without testing by using the texture characteristic features obtained directly from the surface images. As a result, the correlation coefficient values of R2 = 0,97 in air permeability, R2 = 0,90 in breaking strength and R2 = 0,89 in breaking elongation were obtained between experimental results and artificial neural network prediction results.

Photovoltaic maximum power point tracking based on IWD-SVM

Photovoltaic system maximum power point tracking (MPPT) has great potential for improvement of power generation. To optimise MPPT, this paper presents a prediction model based on an intelligent water drops optimisation support vector machine (IWD-SVM) for maximum power point working voltage. The Intelligent Water Drops (IWD) algorithm is used to optimise the penalty factor and kernel function parameters of the SVM, thus improving the training efficiency of the learning machine. Based on the optimisation algorithm, the SVM is used to model the PV array, the prediction results are compared to verify the accuracy and effectiveness of the IWD-SVM model. In addition, the IWD-SVM model is compared with the traditional neural network prediction results, which further verifies the validity of the proposed IWD-SVM model.