Size Of Hidden Layer Research Articles

Optimizing neural network models for predicting nuclear reactor channel temperature: A study on hyperparameter tuning and performance analysis

This study emphasizes how important accurate prediction of channel temperatures in nuclear reactors is for safety and operational efficiency. While traditional methods require long and complex processes such as kernel modeling and mathematical simulations, artificial neural networks (ANN) provide more efficient predictions by accelerating this process. The superior ability of ANNs to process large data sets is intended to demonstrate that this study will provide a valuable alternative compared to conventional methods and increase the accuracy of reactor temperature predictions. In this study, the training performances of Artificial Neural Network (ANN) developed to determine the nuclear reactor channel temperature with different hyperparameter combinations were analysed. It was conducted several experimental studies to assess the influence of hyperparameters on our model for nuclear reactor parameter data prediction. The training and validation results indicates that learning rate, hidden layer sizes and number have critical effects for the more precisive prediction. It was observed that models with a learning rate of 0.05 and 0.5 achieved successful learning with less fluctuation in training and validation errors. When looking at hidden layer sizes, networks with 32 and 64 neurons performed better than networks with 16 neurons. For the test phase our model can successfully predict data with slight error margin. As a result, we demonstrated that neural networks are a powerful tool in nuclear reactor channel temperature prediction through our proposed model.

Analysis of the Housing Market Dynamics Using NARX Neural Network

This study employs a Nonlinear Autoregressive with eXogenous inputs (NARX) neural network to model the dynamics of the housing construction market in Poland, with a distinction made between segments of developers and individual investors. The dataset under analysis contains the 19-year data corresponding to the numbers of housing units approved for construction, under construction, and completed. The NARX model was calibrated thoroughly to suit unique characteristics of the data, with an emphasis put on the hidden layer size and delay parameters, to capture the estate market's nonlinear trends. Results show a very high efficiency of NARX models and highlight distinct patterns and dynamics in the housing completion, construction starts, and permit issuance between the two market segments. These variations are vital for understanding the distinct forces and trends shaping the developers’ and individual investors’ markets in the Polish housing sector. Findings of the analysis provide valuable insight into the nanced functioning of these market segments.

Deep learning-based recognition system for pashto handwritten text: benchmark on PHTI.

This article introduces a recognition system for handwritten text in the Pashto language, representing the first attempt to establish a baseline system using the Pashto Handwritten Text Imagebase (PHTI) dataset. Initially, the PHTI dataset underwent pre-processed to eliminate unwanted characters, subsequently, the dataset was divided into training 70%, validation 15%, and test sets 15%. The proposed recognition system is based on multi-dimensional long short-term memory (MD-LSTM) networks. A comprehensive empirical analysis was conducted to determine the optimal parameters for the proposed MD-LSTM architecture; Counter experiments were used to evaluate the performance of the proposed system comparing with the state-of-the-art models on the PHTI dataset. The novelty of our proposed model, compared to other state of the art models, lies in its hidden layer size (i.e., 10, 20, 80) and its Tanh layer size (i.e., 20, 40). The system achieves a Character Error Rate (CER) of 20.77% as a baseline on the test set. The top 20 confusions are reported to check the performance and limitations of the proposed model. The results highlight complications and future perspective of the Pashto language towards the digital transition.

Efficient ELM Model With Parameter Optimization Using Pso Algorithms in the Prediction of Combustion Pressure Parameters of Dsi Engine Using Ethanol- Gasoline Blends

The present study focused mainly on developing PSO based ELM model to predict cylinder pressure associated parameters. Performance of PSO-ELM model then compared with ELM model to obtain its credential. For training and testing the models, data has been acquired through experiments on a Twin Spark Ignition (TSI) gasoline engine using EGB as fuel. The various operating variables are treated as input data whereas cylinder pressure associated parameters are treated as output data for the model. The result of the proposed modelling study indicated that PSO-ELM model has obtained the best performance with lowest value of MSE, MAPE (%) and hidden layer size as compared to ELM model. Hence PSO-ELM results in an efficient model structure with great generalization performance. Further, it is also observed that PSO-ELM takes more time as it calls for an iterative procedure for searching the optimal solution as compared to ELM, which takes only a single epoch.

IMPLEMENTATION OF TECHNOLOGY FOR IMPROVING THE QUALITY OF SEGMENTATION OF MEDICAL IMAGES BY SOFTWARE ADJUSTMENT OF CONVOLUTIONAL NEURAL NETWORK HYPERPARAMETERS

Background. The scientists have built effective convolutional neural networks in their research, but the issue of optimal setting of the hyperparameters of these neural networks remains insufficiently researched. Hyperparameters affect model selection. They have the greatest impact on the number and size of hidden layers. Effective selection of hyperparameters improves the speed and quality of the learning algorithm. It is also necessary to pay attention to the fact that the hyperparameters of the convolutional neural network are interconnected. That is why it is very difficult to manually select the effective values of hyperparameters, which will ensure the maximum efficiency of the convolutional neural network. It is necessary to automate the process of selecting hyperparameters, to implement a software mechanism for setting hyperparameters of a convolutional neural network. The author has successfully implemented the specified task.
 Objective. The purpose of the paper is to develop a technology for selecting hyperparameters of a convolutional neural network to improve the quality of segmentation of medical images..
 Methods. Selection of a convolutional neural network model that will enable effective segmentation of medical images, modification of the Keras Tuner library by developing an additional function, use of convolutional neural network optimization methods and hyperparameters, compilation of the constructed model and its settings, selection of the model with the best hyperparameters.
 Results. A comparative analysis of U-Net and FCN-32 convolutional neural networks was carried out. U-Net was selected as the tuning network due to its higher quality and accuracy of image segmentation. Modified the Keras Tuner library by developing an additional function for tuning hyperparameters. To optimize hyperparameters, the use of the Hyperband method is justified. The optimal number of epochs was selected - 20. In the process of setting hyperparameters, the best model with an accuracy index of 0.9665 was selected. The hyperparameter start_neurons is set to 80, the hyperparameter net_depth is 5, the activation function is Mish, the hyperparameter dropout is set to False, and the hyperparameter bn_after_act is set to True.
 Conclusions. The convolutional neural network U-Net, which is configured with the specified parameters, has a significant potential in solving the problems of segmentation of medical images. The prospect of further research is the use of a modified network for the diagnosis of symptoms of the coronavirus disease COVID-19, pneumonia, cancer and other complex medical diseases.

Evaluation of drying behavior and characteristics of potato slices in multi–stage convective cabinet dryer: Application of artificial neural network

The inconsistency in the quality of dried products at different coordinates within a conventional multi-stage convective cabinet dryer is a critical but often neglected problem. In this study, the drying behavior (moisture ratio) occurring in each drying tray layer and the drying characteristics (shrinkage or area ratio) occurring at different coordinates within a multi-stage convective cabinet dryer was assessed. Potato slices were used as raw materials in the drying process. Experiments were carried out by varying three different hot air velocities and two different drying temperatures. It was found that under the same hot air temperature and air velocity, the change in moisture content in each drying tray and the shrinkage in each coordinate of the potato slices were different. Artificial neural network model was used to predict the moisture ratio and the area ratio of the potato slices based on the experimental data. The moisture ratio obtained from the experiment was evaluated by comparing it with the drying model. The results showed a good confidence level with the coefficient of determination in the range of 0.962 7–0.993 3. The shrinkage analysis was based on the photographic data taken through image processing before usage as the output data for the predictive model. The predictive model was designed to have various architectures with different parameters; both hidden layer and hidden layer size, learning rate, training cycles, sampling type and split ratio. The best moisture ratio and area ratio model provided the coefficient of determination of 0.996 and 0.970, respectively.

Introducing nonlinear activations into quantum generative models

Due to the linearity of quantum mechanics, it remains a challenge to design quantum generative machine learning models that embed nonlinear activations into the evolution of the state vector. However, some of the most successful classical generative models, such as those based on neural networks, involve highly nonlinear dynamics for quality training. In this paper we explore the effect of these dynamics in quantum generative modeling by introducing a model that adds nonlinear activations via a neural network structure onto the standard Born machine framework---the quantum neuron Born machine (QNBM). To achieve this, we utilize a previously introduced quantum neuron subroutine, which is a repeat-until-success circuit with midcircuit measurements and classical control. After introducing the QNBM, we investigate how its performance depends on network size, by training a three-layer QNBM with four output neurons and various input and hidden layer sizes. We then compare our nonlinear QNBM to the linear quantum circuit Born machine (QCBM). We allocate similar time and memory resources to each model such that the only major difference is the qubit overhead required by the QNBM. With gradient-based training, we show that while both models can easily learn a trivial uniform probability distribution, on a more challenging class of distributions, the QNBM achieves an almost three-times-smaller error rate than a QCBM with a similar number of tunable parameters. We therefore provide evidence that suggests that nonlinearity is a useful resource in quantum generative models, and we put forth the QNBM as a model with good generative performance and potential for quantum advantage.

Intelligent prediction models based on machine learning for CO2 capture performance by graphene oxide-based adsorbents

Designing a model to connect CO2 adsorption data with various adsorbents based on graphene oxide (GO) which is produced from various forms of solid biomass, can be a promising method to develop novel and efficient adsorbents for CO2 adsorption application. In this work, the information of several GO-based solid sorbents were extracted from 17 articles aimed to develop a machine learning based model for CO2 adsorption capacity prediction. The extracted data including specific surface area, pore volume, temperature, and pressure were considered as input parameter, and CO2 uptake capacity was defined as model response, alsoseven different models, including support vector machine, gradient boosting, random forest, artificial neural network (ANN) based on multilayer perceptron (MLP) and radial basis function (RBF), Extra trees regressor and extreme gradient boosting, were employed to estimate the CO2 adsorption capacity. The best performance was obtained for ANN based on MLP method (R2 > 0.99) with hyperparameters of the following: hidden layer size = [45 35 45 45], optimizer = Adam, the learning rate = 0.003, β1 = 0.9, β2 = 0.999, epochs = 1971, and batch size = 32. To investigate CO2 uptake dependency on mentioned effective parameters, three dimensional diagrams were reported based on MLP network, also the MLP network characteristics including weight and bias matrices were reported for further application of CO2 adsorption process design. The accurately predicted capability of the generated models may considerably minimize experimental efforts, such as estimating CO2 removal efficiency as the target based on adsorbent properties to pick more efficient adsorbents without increasing processing time. Current work employed statistical analysis and machine learning to support the logical design of porous GO for CO2 separation, aiding in screening adsorbents for cleaner manufacturing.

Preliminary Modelling of Solar Quiet Geomagnetic Field Average using Non-Linear Autoregressive with Exogeneous Input (NARX)

This paper discusses geomagnetic field attempt modelling using an Artificial Neural Network (ANN). The local horizontal component of geomagnetic field data was collected on April 2011 (equinox) during a solar quiet day at recent solar cycle inclination-24 using the Magnetic Data Acquisition System (MAGDAS) in Langkawi, Malaysia, in the low latitude region. The calculated average values (mean) of the H component geomagnetic field variation during Equinox 2011 characterised the dominant geomagnetic field during that particular solar cycle. The difference in amplitude of maximum and minimum values shows a regular diurnal variation of the geomagnetic field during Sq inthe low latitude region. The output training utilised these calculated mean values during the modelling attempt. Meanwhile, the input training utilised proton density, solar wind plasma speed, plasma flow pressure, and Interplanetary Magnetic Field (IMF) space data using Non-Linear Auto Regressive Input (NARX). The best modelling outputs were 3, 2, and 3 for input delay (𝑛𝑢), output delay (𝑛y), and hidden layer size (h), respectively. Residual test and model fit analysis show an unbiased and high overlap between predicted and actual geomagnetic field average values, suggesting the model can potentially anticipate the geomagnetic field average during solar quiet in November (month of Equinox) on solar cycle inclination.

Prediction of Process Parameters for the Integrated Biomass Gasification Power Plant Using Artificial Neural Network

Alternative renewable fuels like biomass have the potential to be considered for electricity generation by replacing the utilization of fossil fuels and reducing the greenhouse gas emissions into the environment. An integrated biomass gasification power plant is the best suitable option to generate electricity from different biomass feedstocks. Several modeling and simulation techniques have been utilized for the integrated biomass gasification power generation process. These models are utilized to predict the power output from the different gasifier types, designs, and feedstocks. In this study, An Artificial neural network (ANN) model is developed to estimate the process parameters of the Integrated biomass gasification power plant. This ANN model predicts the gasification temperature (T) and air to fuel ratio (AFR) for the gasification process integrated with the power plant at the atmospheric pressure. There is a total of ten input parameters such as moisture content of biomass (M), volatile matter (VM), fixed carbon (FC), ash content (A), element composition of carbon (C), oxygen (O), hydrogen (H), nitrogen (N), sulfur (S) and required power (KW) are used to predict the two key gasification process parameters T and AFR. The data generated from thermodynamic equilibrium model simulations are employed in the developed ANN model for the different 86 biomass feedstocks. The proposed ANN model was optimized for the Mean Squared Error (MSE) loss function and evaluated using MSE and R score metrics. It is observed that the best predicted for a hidden layer size was of 60 neurons. The best test score was achieved as an MSE score of 1,497 and test R 0.9976. This study can be implemented for any kind of biomass feedstock for the power generation system.

Automated Delineation of Supraglacial Debris Cover Using Deep Learning and Multisource Remote Sensing Data

High-mountain glaciers can be covered with varying degrees of debris. Debris over glaciers (supraglacial debris) significantly alter glacier melt, velocity, ice geometry, and, thus, the overall response of glaciers towards climate change. The accumulated supraglacial debris impedes the automated delineation of glacier extent owing to its similar reflectance properties with surrounding periglacial debris (debris aside the glaciated area). Here, we propose an automated scheme for supraglacial debris mapping using a synergistic approach of deep learning and multisource remote sensing data. A combination of multisource remote sensing data (visible, near-infrared, shortwave infrared, thermal infrared, microwave, elevation, and surface slope) is used as input to a fully connected feed-forward deep neural network (i.e., deep artificial neural network). The presented deep neural network is designed by choosing the optimum number and size of hidden layers using the hit and trial method. The deep neural network is trained over eight sites spread across the Himalayas and tested over three sites in the Karakoram region. Our results show 96.3% accuracy of the model over test data. The robustness of the proposed scheme is tested over 900 km2 and 1710 km2 of glacierized regions, representing a high degree of landscape heterogeneity. The study provides proof of the concept that deep neural networks can potentially automate the debris-covered glacier mapping using multisource remote sensing data.

Developmental and evolutionary constraints on olfactory circuit selection

SignificanceIn this work, we explore the hypothesis that biological neural networks optimize their architecture, through evolution, for learning. We study early olfactory circuits of mammals and insects, which have relatively similar structure but a huge diversity in size. We approximate these circuits as three-layer networks and estimate, analytically, the scaling of the optimal hidden-layer size with input-layer size. We find that both longevity and information in the genome constrain the hidden-layer size, so a range of allometric scalings is possible. However, the experimentally observed allometric scalings in mammals and insects are consistent with biologically plausible values. This analysis should pave the way for a deeper understanding of both biological and artificial networks.

Comparison of sensitivity matrix method, power function-based response surface method, and artificial neural network in the analysis of building fire egress performance

Unlike the areas of building energy and structural health, performance monitoring tools are currently absent in the area of building fire protection. Computational Fluid Dynamics (CFD) models like Fire Dynamics Simulator (FDS) are widely applied in building fire performance design, which can be equally used to predict changes of building fire performance. However, due to its time-consuming nature, it is not realistic to apply FDS frequently. The sensitivity matrix method (SMM) has been discussed as a quick method to predict changes of building fire egress performance. However, this approach can have significant uncertainties when being applied to datasets with many input parameters due to its inherent incapability of predicting accurately system responses if input data are considerably far away from the baseline points around which a SMM is developed. Response surface methods (RSM) are commonly used to characterize the relationships between input variables and output quantities for complicated problems. Different from conventional RSMs, a novel two phase power function fitting process is proposed to develop substitute algebraic models of the available safe egress time (ASET) from FDS numerical experiments based on a theorem which states that an output variable is proportional to the product of input parameters to their respective powers if the output variable is proportional to each input parameter to some power and the input parameters are independent of each other. An artificial neural network (ANN) is a universal method to approximate any arbitrary complicated, nonlinear system response with limited number of discontinuities without deep understanding of how the system works. This paper employs MATLAB's feedforward neural networks with error backpropagating algorithm to approximate the FDS response. Applicability in terms of uncertainties including system bias and relative standard deviation (RSD) or percentage of predictions falling in a preset acceptable error range are compared among RSMs developed from various datasets, ANNs with various hidden layer sizes and dataset sizes, and SMMs which use the same fire scenarios in a small three-story apartment building. The result shows that it is possible for ANNs to have lowest model uncertainties and highest percentage of predictions within the preset 20% error scope as far as the specific fire egress safety problem discussed in this paper is concerned, but the cost of developing a SMM, namely the number of data cases, is the lowest. Due to the different aspects of RSMs, ANNs, and SMMs, to better understand the building fire performance gap continuously, a hybrid strategy of starting with SMMs followed by RSMs and/or ANNs is recommended in a fire performance monitoring tool. • A novel Response Surface Method is proposed based on two-phase power function fitting. • MATLAB's feedforward NN with error backpropagating algorithm is employed to approximate the FDS simulation results. • The applicability of Sensitivity Matrix Method, Response Surface Method, and Artificial Neural Network is compared. • A hybrid strategy of starting with SMM followed by RSM and/or ANN is recommended in a fire performance monitoring tool.

Forecast for Market Clearing Price with Artificial Neural Networks in Day Ahead Market

In this study, the Market Clearing Price (MCP) is forecasted with Artificial Neural Networks and the modeling success is examined for different preprocessing strategies. The purpose of the study is to obtain the optimum model with a significant estimation success and to provide the best price prediction. The hour-based electricity generation data of diverse production items are assigned as inputs and the resulting MCP is modeled. The raw data are first cleaned from outliers, then subjected to different normalization processes and 70 different ANNs are trained. Additionally, networks are trained with data classified in seasons and the effect of seasonal patterns on model success is observed. Finally, networks showing the optimum performance are selected. It is noted that the type of the normalization strategy and the hidden layer size are the key factors to make a decent estimation. Then, in order to test the networks with extreme cases, data for the special days (official holidays) are applied to these networks as input. The success of the networks is evaluated by comparing the MCP predictions with the actual values. It is revealed to make a prediction for official holidays, a model which is special to this period of year is required.

DATA MINING DENGAN ALGORITMA NEURAL NETWORK DAN VISUALISASI DATA UNTUK PREDIKSI KELULUSAN MAHASISWA

Graduating on time is the desire of all students, not only that graduating on time is an advantage for both parties, namely students and educational institutions. But the graduation status of students when predicted does not always produce predictions early, so that it can result in graduation not on time. This research was conducted using data of students who graduated for 4 years from 2016-2019. The classification method is an approach to grouping data in data mining, namely classifying data. This classification method can also be used to make predictions for information that has not been previously known. The classification data mining method that will be used is the neural network algorithm. Visualization to display recapitulation data visually more interesting and neural network algorithm to predict student graduation which is difficult to do manually. Attributes used in training data consist of Jenis Kelamin, Asal, Kelas, Jurusan, Umur, IPK, Tanggal Yudicium, Tahun Yudicium and Class Hasil. The attributes that become parameters are 9 attributes, of which 8 are predictor attributes and 1 are results attributes. Training and testing data by changing parameters, namely: Hidden Layer Size: 3, Training Time: 500, Learning Rate: 0.3, Momentum: 0.2 produces a classification showing the level of accuracy using a neural network algorithm is 92.83%. Displays some very complete reporting recapitulation, so that predictions and visualization of the data can help in graduating students and provide recommendations for appropriate actions and must be done by management or the authorities to make decisions.

A BAYESIAN APPROACH BASED ON ACQUISITION FUNCTION FOR OPTIMAL SELECTION OF DEEP LEARNING HYPERPARAMETERS: A CASE STUDY WITH ENERGY MANAGEMENT DATA

With the recent rollout of smart meters, huge amount of data can be generated on hourly and daily basis. Researchers and industry persons can leverage from this big data to make intelligent decisions via deep learning (DL) algorithms. However, the performance of DL algorithms are heavily dependent on the proper selection of parameters. If the hyperparameters are poorly selected, they usually lead to suboptimal results. Traditional approaches include a manual setting of parameters by trial and error methods which is time consuming and difficult process. In this paper, a Bayesian approach based on acquisition is presented to automatic selection of optimal parameters based on provided data. The acquisition function was established to search for the best parameter from the input space and evaluate the next points based on past observations. The tuning process identifies the best model parameters by iterating the objective function and minimizing the loss for optimizable variables such as learning rate and Hidden layersize. To validate the presented approach, we conducted a case study on real-life energy management datasets while constructing a deep learning model on MATLAB platform. A performance comparison was drawn with random parameters and optimal parameters selected by presented approach. The comparison results illustrate that the presented approach is effective as it brings a notable improvement in the performance of learning algorithm.

Price forecast of Yunnan Pu’er tea based on the MLP neural network

Pu’er tea as the landmark product of Yunnan province, which plays a decisive role in the national tea industry and the economic development of Yunnan province. At present, the qualitative and quantitative methods cannot be accurately predicted the price of Pu’er tea. In order to solve this problem, this paper tries to use MLP neural network to forecast Pu’er tea price. Firstly, determine the influencing factors of Pu’er tea price, then established the MLP neural network model, selection activation function and design the number of neurons and the hidden layer size, finally compare the method between MLP and random forest to verify the feasibility and effectiveness of the MLP method.

Instar Determination of Blaptostethus pallescens (Hemiptera: Anthocoridae) Using Artificial Neural Networks

Abstract Blaptostethus pallescens Poppius is an important predator of vegetable pests in tropical regions. The correct identification of the stages of the life cycle of predatory species is crucial, since different stages may present different rates of pest consumption. Artificial neural networks (ANNs) are computational tools with a structure based on the human brain. With applications in several fields, ANNs have been applied in pest management for identification of pest species, spatial distribution modeling, and insect forecasting. The objective of this study was to apply ANNs as a method for the instar determination of B. pallescens using three morphometric measures (head width, body width, and body length). Cluster analysis was performed to categorize the insects in instars according to the morphometric variables. Subsequently, the ANNs were trained for instar determination using the morphometric measures as input variables. The ANNs tested (with 2, 4, 6, 8, 10, and 12 hidden neurons) provided proper data fitting (R2 > 98%). However, due to the parsimony principle, the network with hidden layer size 6 was selected. This study shows the successful application of ANNs in the instar determination of B. pallescens, which would not be possible using classical methods.

An Investigation of Dimensionality Reduction Techniques for EMG-based Force Estimation.

In this paper, extracted features in time and frequency domain, from high-density surface electromyogram (HD-sEMG) signals acquired from the long head and short head of biceps brachii, and brachioradialis during isometric elbow flexion are used to estimate force induced at the wrist using an artificial neural network (ANN). Different hidden layer sizes were considered to investigate its effect on the model accuracy. Also, we applied two dimensionality reduction techniques, principal component analysis (PCA) and t-distributed stochastic neighbor embedding (t-SNE), on the feature set and investigated their effects on force estimation accuracy.

Adaptive CCR-ELM with variable-length brain storm optimization algorithm for class-imbalance learning

Class-specific cost regulation extreme learning machine (CCR-ELM) can effectively deal with the class imbalance problems. However, its key parameters, including the number of hidden nodes, the input weights, the biases and the tradeoff factors are normally generated randomly or preset by human. Moreover, the number of input weights and biases depend on the size of hidden layer. Inappropriate quantity of hidden nodes may lead to the useless or redundant neuron nodes, and make the whole structure complex, even cause the worse generalization and unstable classification performances. Based on this, an adaptive CCR-ELM with variable-length brain storm optimization algorithm is proposed for the class imbalance learning. Each individual consists of all above parameters of CCR-ELM and its length varies with the number of hidden nodes. A novel mergence operator is presented to incorporate two parent individuals with different length and generate a new individual. The experimental results for nine imbalance datasets show that variable-length brain storm optimization algorithm can find better parameters of CCR-ELM, resulting in the better classification accuracy than other evolutionary optimization algorithms, such as GA, PSO, and VPSO. In addition, the classification performance of the proposed adaptive algorithm is relatively stable under varied imbalance ratios. Applying the proposed algorithm in the fault diagnosis of conveyor belt also proves that ACCR-ELM with VLen-BSO has the better classification performances.