Number Of Neurons In Hidden Layer Research Articles

Prediction of the Moment Capacity of FRP-Strengthened RC Beams Exposed to Fire Using ANNs

In this study, an Artificial Neural Network (ANN) model to predict the moment capacity of Fiber Reinforced Plastic (FRP) strengthened Reinforced Concrete (RC) beams exposed to fire is developed. The software ABAQUS heat transfer analysis is verified by comparison with the fire resistance test results. Through this heat transfer analysis, the temperature distribution of the beam section is determined, and 400 datasets are obtained using the moment capacity calculation method combined with the section equilibrium method. The data consist of eight input parameters: the beam width, beam height, FRP area, rebar area, concrete compressive strength, insulation thickness, concrete cover depth, and fire exposure time. The output parameter is the moment capacity. The ANN model is developed through a sensitivity study using the algorithm type and the number of hidden-layer neurons as variables. The average error between the predicted data of the developed ANN model and the target data obtained from the moment capacity calculation method was 0.35 kN·m, and the average relative error was 0.2512%, showing high accuracy. Therefore, the ANN model developed here can determine the moment capacity without complex calculations. The effects of the input parameters on the moment capacity of the FRP-strengthened RC beams exposed to fire are investigated using the ANN model.

Load Forecasting Based on LVMD-DBFCM Load Curve Clustering and the CNN-IVIA-BLSTM Model

Power load forecasting plays an important role in power systems, and the accuracy of load forecasting is of vital importance to power system planning as well as economic efficiency. Power load data are nonsmooth, nonlinear time-series and “noisy” data. Traditional load forecasting has low accuracy and curves not fitting the load variation. It is not well predicted by a single forecasting model. In this paper, we propose a novel model based on the combination of data mining and deep learning to improve the prediction accuracy. First, data preprocessing is performed. Second, identification and correction of anomalous data, normalization of continuous sequences, and one-hot encoding of discrete sequences are performed. The load data are decomposed and denoised using the double decomposition modal (LVMD) strategy, the load curves are clustered using the double weighted fuzzy C-means (DBFCM) algorithm, and the typical curves obtained are used as load patterns. In addition, data feature analysis is performed. A convolutional neural network (CNN) is used to extract data features. A bidirectional long short-term memory (BLSTM) network is used for prediction, in which the number of hidden layer neurons, the number of training epochs, the learning rate, the regularization coefficient, and other relevant parameters in the BLSTM network are optimized using the influenza virus immunity optimization algorithm (IVIA). Finally, the historical data of City H from 1 January 2016 to 31 December 2018, are used for load forecasting. The experimental results show that the novel model based on LVMD-DBFCM load c1urve clustering combined with CNN-IVIA-BLSTM proposed in this paper has an error of only 2% for electric load forecasting.

Prediction of pyrolysis product yield of medical waste based on BP neural network

Predicting pyrolysis product yields becomes more and more important for the realization of harmless, reduction, and resource disposal of medical waste. Based on the gray correlation analysis, this study predicted the three-state yield of medical waste pyrolysis products by building a BP neural network model. A database of pyrolysis product yields distribution of multivariate organic solid waste was established by selecting 225 sets of literature data in the existing research on rapid pyrolysis of organic solid waste and 18 sets of experimental data obtained from independent experiments. By the combination of the S-type tansig function of the hidden layer with the linear transfer purelin function of the output layer, the transfer function was created to construct the mapping between input and output variables. The results showed that when the number of hidden layer neurons was 14 and the learning rate was 0.04, the predicted yields and experimental yields were in good agreement with the R2 of neural network training and testing samples were 0.943 and 0.889, respectively. Increasing temperature and prolonging solid residence time were beneficial to the conversion of solid phase products to gas and liquid phase products. The effect of gas residence time on the three-phase yield distribution was less than that of temperature and solid residence time.

Performance Assessment of Joint Optical-Digital Nonlinearity Mitigation Schemes in Long-Haul Systems

The combined approach of optical and digital nonlinearity mitigation techniques have been shown to have an edge over the individual schemes, tackling their drawbacks and improving compensation capabilities. We demonstrated fiber-nonlinearity mitigation of 32 GBd single-polarization 16QAM signals transmitted over an 800-km dispersion-managed link using optical phase conjugation (OPC) and digital-domain neural networks (NN). Our implemented NN comprised a recurrent neural network (RNN), and a reservoir computing network (RCN). To further compensate for the penalty introduced by the design of OPC, we employed NN schemes in the digital signal processing and applied it to the signal after transmission in the OPC-assisted link. We present the results for optimizing the NN models for important hyper-parameters like the number of hidden layer neurons and the input vector size. The proposed joint approach achieves Q <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> -factor improvements up to 1.8 dB while surpassing the improvements of the individual schemes. Our experiments indicate that the joint approach has the potential to reduce the overall complexity of NN architectures in terms of the size of the hidden layer and input vector.

E-Commerce Enterprises Financial Risk Prediction Based on FA-PSO-LSTM Neural Network Deep Learning Model

The rapid development of Internet information technology has made e-commerce enterprises face complex and changing financial problems. Combining artificial intelligence algorithms and dynamic monitoring of financial risks has been a current research hotspot. Based on this, this paper conducts an empirical study with a sample of listed Chinese e-commerce enterprises from 2012 to 2022. Firstly, using factor analysis (FA) to obtain the common factors between the original financial and non-financial indicators has the effect of reducing the overfitting risk of the model. Secondly, the mean square error (MSE) of the output and predicted values of the Long Short-Term Memory neural network (LSTM) is used as the fitness function of the intelligent swarm optimization algorithm, and then the Particle Swarm Optimization (PSO) algorithm is used to optimize the learning rate (LR) and the number of hidden layer neurons in the Long Short-Term Memory (LSTM) neural network. Finally, a financial risk prediction model based on FA-PSO-LSTM deep learning is constructed, and multiple benchmark models are introduced for comparative analysis on each evaluation index. The study shows that for nonlinear multivariate data with complex structure, the fused deep learning model proposed in this paper achieves the lowest values in mean square error (MSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). This indicates that the model has the best prediction effect, which is helpful to help managers make relevant decisions efficiently and scientifically and make the enterprise sustainable.

Robust state of charge estimation of lithium-ion battery via mixture kernel mean p-power error loss LSTM with heap-based-optimizer

The state of charge (SOC) estimation of lithium-ion battery is an important function in the battery management system (BMS) of electric vehicles. The long short term memory (LSTM) model can be employed for SOC estimation, which is capable of estimating the future changing states of a nonlinear system. Since the BMS usually works under complicated operating conditions, i.e the real measurement data used for model training may be corrupted by non-Gaussian noise, and thus the performance of the original LSTM with the mean square error (MSE) loss may deteriorate. Therefore, a novel LSTM with mixture kernel mean p-power error (MKMPE) loss, called MKMPE-LSTM, is developed by using the MKMPE loss to replace the MSE as the learning criterion in LSTM framework, which can achieve robust SOC estimation under the measurement data contaminated with non-Gaussian noises (or outliers) because of the MKMPE containing the p-order moments of the error distribution. In addition, a meta-heuristic algorithm, called heap-based-optimizer (HBO), is employed to optimize the hyper-parameters (mainly including learning rate, number of hidden layer neuron and value of p in MKMPE) of the proposed MKMPE-LSTM model to further improve its flexibility and generalization performance, and a novel hybrid model (HBO-MKMPE-LSTM) is established for SOC estimation under non-Gaussian noise cases. Finally, several tests are performed under various cases through a benchmark to evaluate the performance of the proposed HBO-MKMPE-LSTM model, and the results demonstrate that the proposed hybrid method can provide a good robustness and accuracy under different non-Gaussian measurement noises, and the SOC estimation results in terms of mean square error (MSE), root MSE(RMSE), mean absolute relative error (MARE), and determination coefficient R2 are less than 0.05%, 3%, 3%,and above 99.8% at 25℃, respectively.

Developing a Tuned Three-Layer Perceptron Fed with Trained Deep Convolutional Neural Networks for Cervical Cancer Diagnosis.

Cervical cancer is one of the most common types of cancer among women, which has higher death-rate than many other cancer types. The most common way to diagnose cervical cancer is to analyze images of cervical cells, which is performed using Pap smear imaging test. Early and accurate diagnosis can save the lives of many patients and increase the chance of success of treatment methods. Until now, various methods have been proposed to diagnose cervical cancer based on the analysis of Pap smear images. Most of the existing methods can be divided into two groups of methods based on deep learning techniques or machine learning algorithms. In this study, a combination method is presented, whose overall structure is based on a machine learning strategy, where the feature extraction stage is completely separate from the classification stage. However, in the feature extraction stage, deep networks are used. In this paper, a multi-layer perceptron (MLP) neural network fed with deep features is presented. The number of hidden layer neurons is tuned based on four innovative ideas. Additionally, ResNet-34, ResNet-50 and VGG-19 deep networks have been used to feed MLP. In the presented method, the layers related to the classification phase are removed in these two CNN networks, and the outputs feed the MLP after passing through a flatten layer. In order to improve performance, both CNNs are trained on related images using the Adam optimizer. The proposed method has been evaluated on the Herlev benchmark database and has provided 99.23 percent accuracy for the two-classes case and 97.65 percent accuracy for the 7-classes case. The results have shown that the presented method has provided higher accuracy than the baseline networks and many existing methods.

Modeling triangular, rectangular, and parabolic weirs using weighted robust extreme learning machine

In this study, dimensionless parameters influencing the coefficient of discharge (COD) are found and four different WRELM models are developed. After that, a dataset is created for verifying the WRELM models in which 70% of the data are employed to train learning machine models and the remaining 30% to test them. For the mentioned algorithm, the optimal number of hidden layer neurons along with the best activation function is chosen. Additionally, the best value for the regularization parameter of the WRELM algorithm is computed. By analyzing the simulation results, the superior WRELM model and the variables impacting the COD are detected. The superior WRELM model approximates COD values with the minimum error and the highest correlation with laboratory values. For the superior model, the values of the R, MAE and VAF statistical indices are computed to be 0.994, 0.0004 and 98.662, respectively. The analysis of the simulation results indicates that the dimensionless parameters α and T/B are the most influencing input parameters. The superior WRELM model results are compared with the algorithm, and it is concluded that the WRELM model is noticeably more efficient. For the superior WRELM model, a partial derivative sensitivity analysis (PDSA) is conducted in which as the input parameter α increases, the PSDA value increases as well. Finally, an equation is suggested for estimating COD values.

Analysis of Factors Affecting Import and Export of Underdeveloped Regions Based on RBF Neural Network - A Case Study of Guizhou Province

Guizhou Province, as an inland province of China without borders and seas, has relatively backward economic development, but its economic growth is fast and its development potential is great. Studying its import and export trade is important to improve residents ' income and accelerate internationalization. This paper analyzes and establishes the indicator system of regional import and export influencing factors through literature analysis, and then uses ( radial basis function ) RBF neural network to analyze the influencing factors of import and export in Guizhou Province. Taking the economic data of Guizhou Province from 2001 to 2020 as samples, RBF neural network is used as a tool to analyze the relationship between various influencing factors and total import and export volume. The results show that when the number of hidden layer neurons is 14 or more, the RBF neural network model accuracy is the highest, R2 is 0.7005, and MAPE is 0.0834. Among the first-level indicators affecting import and export, the importance of the regional internal economic level is the highest. Among the second-level indicators, the importance of scientific and technological progress is the highest, followed by GDP.

Comparison of Artificial Neural Network and Polynomial Approximation Models for Reflectance Spectra Reconstruction

Knowledge of surface reflection of an object is essential in many technological fields, including graphics and cultural heritage. Compared to direct multi- or hyper-spectral capturing approaches, commercial RGB cameras allow for a high resolution and fast acquisition, so the idea of mapping this information into a reflectance spectrum (RS) is promising. This study compared two modelling approaches based on a training set of RGB-reflectance pairs, one implementing artificial neural networks (ANN) and the other one using multivariate polynomial approximation (PA). The effect of various parameters was investigated: the ANN learning algorithm-standard backpropagation (BP) or Levenberg-Marquardt (LM), the number of hidden layers (HLs) and neurons, the degree of multivariate polynomials in PA, the number of inputs, and the training set size on both models. In the two-layer ANN with significantly fewer inputs than outputs, a better MSE performance was found where the number of neurons in the first HL was smaller than in the second one. For ANNs with one and two HLs with the same number of neurons in the first layer, the RS reconstruction performance depends on the choice of BP or LM learning algorithm. RS reconstruction methods based on ANN and PA are comparable, but the ANN models' better fine-tuning capabilities enable, under realistic constraints, finding ANNs that outperform PA models. A profiling approach was proposed to determine the initial number of neurons in HLs-the search centre of ANN models for different training set sizes.

Optimization Application of Deep Belief Network in Gear Fault Diagnosis

Deep belief networks (DBNs) in the process of the application of the gear vibration signal fault diagnosis, vector network depth, the number of neurons in hidden layer, set the initial parameters, such as there is a certain blindness, and for different data sets, the influence of initial parameters is different also, in order to get ideal fault recognition rate, the initial parameters have to be repeatedly debugging, This process is not only time-consuming and laborious, but also disadvantageous to its universal application in fault diagnosis. To solve this problem, this paper mainly optimizes the number of hidden layer neurons and learning rate of DBNs. The diagnosis example shows that after optimization, the initial parameter setting is more accurate and efficient, and the fault identification rate is significantly improved.

Research of GPCEMBP Glucose Prediction Algorithm Based on Continuous Glucose Monitoring

In view of the poor timeliness of dynamic blood glucose data and a delay of insulin effect for blood glucose control, and considering the nonlinearity and nonstationarity of the glucose data, a new blood Glucose Prediction algorithm combined Correlation coefficient‐based complete ensemble empirical mode decomposition with adaptive noise and back propagation neural network (GPCEMBP) was proposed to increase the prediction time and improve the prediction accuracy. It refined the mode decomposition algorithm and integrated the correlation mode filter function to extract the characteristic intrinsic mode functions from the original signal. A new neural network prediction model was constructed by optimizing the number of hidden layer neurons, the number of hidden layers, activation functions, the number of inputs, structure, and other parameters. Finally, a predicted blood glucose value was calculated by phase space reconstruction technology. Through ablation and comparison experiments, it was demonstrated that the GPCEMBP algorithm had better prediction accuracy, convergence, and robustness in blood glucose prediction within 84 min. In addition, it has good adaptability to deal with different quality glucose data.

Artificial Neural Networks for the Prediction of Monkeypox Outbreak.

While the world is still struggling to recover from the harm caused by the widespread COVID-19 pandemic, the monkeypox virus now poses a new threat of becoming a pandemic. Although it is not as dangerous or infectious as COVID-19, new cases of the disease are nevertheless being reported daily from many countries. In this study, we have used public datasets provided by the European Centre for Disease Prevention and Control for developing a prediction model for the spread of the monkeypox outbreak to and throughout the USA, Germany, the UK, France and Canada. We have used certain effective neural network models for this purpose. The novelty of this study is that a neural network model for a time series monkeypox dataset is developed and compared with LSTM and GRU models using an adaptive moment estimation (ADAM) optimizer. The Levenberg-Marquardt (LM) learning technique is used to develop and validate a single hidden layer artificial neural network (ANN) model. Different ANN model architectures with varying numbers of hidden layer neurons were trained, and the K-fold cross-validation early stopping validation approach was employed to identify the optimum structure with the best generalization potential. In the regression analysis, our ANN model gives a good R-value of almost 99%, the LSTM model gives almost 98% and the GRU model gives almost 98%. These three model fits demonstrated that there was a good agreement between the experimental data and the forecasted values. The results of our experiments show that the ANN model performs better than the other methods on the collected monkeypox dataset in all five countries. To the best of the authors' knowledge, this is the first report that has used ANN, LSTM and GRU to predict a monkeypox outbreak in all five countries.

An adaptive artificial neural network for reliability analyses of complex engineering systems

Reliability analyses of engineering systems require methods capable of reducing the number of function calls and response variances to calculate the failure probability. This paper is aimed to employ a combination of the Monte Carlo simulation (MCS), artificial neural network (ANN) and control variate technique (CVT) to improve the efficiency and accuracy of the reliability analyses of various engineering systems. The ANN used in this paper is of the multilayer perceptron type. The paper uses the design of experiments (DoE), importance sampling and uniform sampling methods to generate the samples. The combination of the MCS, ANN model, DoEs, and CVT is used for the reliability analyses and significantly improved the results in terms of efficiency. Various numerical examples have shown the superiority of the improved ANN model using CVT in reducing the number of calls of the limit state function and revealed that the obtained results are not sensitive to the number of hidden layer neurons. The reduced number of required simulations and computation costs has enabled the proposed method to be applicable to reliability analyses in complex engineering systems such as structures made by functionally graded materials.

Attention-Based BiLSTM Model for Pavement Temperature Prediction of Asphalt Pavement in Winter

Pavement temperature is the main factor determining road icing, and accurate and timely pavement temperature prediction is of significant importance to regional traffic safety management and preventive maintenance. The prediction of pavement temperature at the micro-scale has been a challenge to be tackled. To solve this problem, a bidirectional extended short-term memory network model based on the attention mechanism (Att-BiLSTM) was proposed to improve the prediction performance by using the time series features of pavement temperature and meteorological factors. Pavement temperature data and climatic data were collected from a road weather station in Yunnan, China. The results show that the MAE, MSE, and MAPE of the proposed Att-BiLSTM model were 0.330, 0.339, and 10.1%, respectively, which were better than the other baseline models. It was shown that 93.4% of the predicted values had an error less than 1 °C, and 82.1% had an error less than 0.5 °C, indicating that the proposed Att-BiLSTM model enables significant performance improvement. In addition, this paper quantified and analyzed the effects of parameters such as the size of the sliding window, the number of hidden layer neurons, and the optimizer on the performance of the prediction model.

Evaluation of neural network parameters in the prediction of AISI 1050 steel machining performance

ABSTRACT Significant improvement in the design of ANN model has been hindered by the lack of evaluation of alternate model parameters. It is important that in generating a suitable ANN model, the best parameters capable of providing the fastest network and the most accurate test result be employed. In this study, the performance of the NN models based on the most suitable NN parameters such as the number of hidden layer neurons ranging from 2–20, activation functions; sigmoid (logsig), tangent sigmoid (tansig), and the linear (purelin), along with three different training algorithm (LM, SCG, and RP), has been evaluated to accurately predict the response values; metal removal rate (MRR) and tool wear in the turning operation of AISI 1050 steel bar. A great difference was observed in the performance of the activation functions and in the training algorithms. The optimal ANN model prediction, recorded R values of 0.827, 0.920 and 0.983 for the training, validation and testing sub-datasets respectively, while an overall R-value of 0.876 was obtained for the training process, signifying confirmation of the predictive ability of the model. The experiments demonstrate that the identification of the most suitable activation functions can lead to significant performance improvements without actually increasing the number of neurons.

A novel neural network and grey correlation analysis method for computation of the heat transfer limit of a loop heat pipe (LHP)

A loop heat pipe (LHP) has the advantages of larger heat transfer capacity and anti-gravity operational performance. The current prediction models for LHP heat transfer capacity have the difficulties in popularization of data volume and determination of accurate parametrical data, leading to the uncertain and varying outcomes that are inconsistent and away from reality. To address these challenges, this paper developed a first-of-its-kind big-data-driven LHP heat transfer limit prediction model by employing the neural network and grey correlation analysis method, which have advantages of high precision and large data volume. A double-layer feedforward neural network with sigmoid hidden neuron and linear output neuron was constructed to predict the heat transfer limit of the LHP. The grey scale analysis is applied to select the variables with correlation coefficient greater than 0.5, thus giving the clear identification of the both input parameters (e.g. refrigerant temperature, filling liquid quantity, height difference between evaporator and condenser, and number of heat pipe array) and output ones (heat transfer limit). The previously validated LHP heat transfer limit calculation model is used to calculate the heat transfer limit corresponding to the selected parameters, thus formulating 1,010,038 sets of data points. Of those calculated datasets, 707,026 (70% of data) are treated as a training set, 151,506 (15% of data) as a verification set, and 151,506 groups of data (15% of data) as the test sets for training. After several optimization and debugging, the number of hidden layer neurons is determined to be 100. The correlation coefficient (R), mean square error (MSE) and mean relative error (MRE) are 0.9997, 52.7 and 0.32% respectively, all of which are within reasonable accuracy range. The results show that the model has good prediction accuracy and consistence and is an effective tool to characterize and optimize the LHP in various application synergies.

Research on State Trend Prediction of Hydraulic Turbine Units Based on Mult-dimensional and Multi-condition Data

Aiming at the problem that it is difficult to accurately predict the flexible operation conditions and state trend of hydraulic turbine, in this paper a cascade model of BP-LSTM classification prediction is proposed, which can identify the working conditions of existing fusion data, and then predict the measuring points of different working conditions.<BR> Based on the pressure parameters of hydraulic turbine units, the improved BP neural network is used to determine the operation conditions of hydraulic turbine units, and the classified data is redivided to establish the multivariate LSTM prediction model. By optimizing the parameters of the multivariate LSTM prediction model, such as the structure, the number of network layers and the number of hidden layer neurons, finally established the cascade model of BP-LSTM classification prediction of time series of hydraulic turbine units.<BR> Through experimental verification and analysis, BP-LSTM classification prediction model can predict the operation trend of measuring points under different working conditions after classification. Compared with other models, BP-LSTM model has higher prediction accuracy and better effect. The cascade model of BP-LSTM classification prediction of time series provides a model basis for the research of predictive control of hydraulic turbine units.

An integrated framework of gated recurrent unit based on improved sine cosine algorithm for photovoltaic power forecasting

Accurate prediction of photovoltaic power is of great significance to the storage and utilization of solar power. In this research, a deep learning model for photovoltaic power prediction based on gated recurrent unit network (GRU), improved sine cosine algorithm (ISCA), and complete ensemble empirical mode decomposition (CEEMD) is proposed. Firstly, CEEMD is used to decompose the original photovoltaic data into several intrinsic mode function (IMF) components and one residual. Secondly, each sub-pattern after decomposition is processed by partial least-squares analysis (PLS). Third, the nonlinear strategy is used to improve SCA, and the Hill-climbing strategy is added to the local search part to improve the performance of the algorithm. Fourth, each sub-pattern is predicted by GRU, then the learning rate and the number of hidden layer neurons of GRU are optimized by the ISCA. Finally, the predicted results of each sub-model are combined to generate the final prediction results. In this study, the proposed model is applied to four photovoltaic power data sets, and different experimental comparison models are established. The experimental results show that the CEEMD-PLS-ISCA-GRU model in this study can obtain good prediction results in all data sets.

Quality prediction and control of thin-walled shell injection molding based on GWO-PSO, ACO-BP, and NSGA-II

Abstract ECG recorders are precision medical devices, but their thin-walled shells are susceptible to warpage and shrinkage during injection molding production due to the injection molding process, which greatly shortens their service life. To address this problem, a multiobjective optimization method for injection molding process parameters based on a combination of a BP neural network model optimized by an ant colony algorithm (ACO-BP) and an improved non-dominated sorting genetic algorithm (NSGA-II) is proposed. The study takes the warpage deformation amount and volume shrinkage rate of the plastic part as the optimization objectives, and the melt temperature, mold temperature, injection pressure, holding pressure, holding time, and cooling time as the design variables. However, for BP neural networks, it is crucial to choose an appropriate number of hidden layer neurons, so the particle swarm algorithm combined with the grey wolf algorithm (GWO-PSO) is used to solve for the optimal number of hidden layer neurons. Firstly, the number of hidden layer neurons of the BP network model was solved based on the samples obtained from the Box–Behnken experimental design and the GWO-PSO algorithm, and the ACO-BP algorithm was used to build the prediction models for warpage and volume shrinkage, respectively, and then combined with NSGA-II for global optimisation. The pareto optimal solution set was subjected to CRITIC analysis and the optimal process parameters were finally obtained, with a minimum warpage of 0.3293 mm and minimum volume shrinkage of 4.993%, a reduction of 8.93 and 6.95% respectively compared to the pre-optimisation period. At the same time, injection molding tests were carried out on the optimum process parameters, and it was found that the molding quality of the plastic parts was better and met the actual production requirements through measurement. The research in this paper provides a theoretical basis for further improving the quality defects of the thin-walled injection molded parts.