Multilayer Perception Neural Network Research Articles

Prediction of soil wind erodibility using a hybrid Genetic algorithm – Artificial neural network method

Understanding and consequent modeling of soil wind erodibility is hampered by the complex nature of the eroding processes and limited empirical data. It is often necessary to resort to robust approaches capable of finding correlated patterns among soil erodibility magnitudes and their drivers. To signify soil erodibility to wind, we used a portable wind tunnel to measure wind erosion rate (g m−2 s−1) at a total of 118 sites in Kerman Province, southeast Iran. At each sampling site, 17 different factors affecting soil erodibility were measured. Gravel coverage, surface crust, very fine and very coarse sands, aggregate stability, and calcium carbonate equivalent (CCE) were introduced as the more important parameters affecting soil erodibility by hybrid Genetic Algorithm-Artificial Neural Network (GA-ANN). A Multi-Layer Perception (MLP) neural network was developed to predict erodibility changes in response to spatial variation of the selected features. The developed MLP-model provided a strong basis for the prediction of soil erodibility, where the coefficient of determination (R2) values of 0.89 and 0.87 were obtained by comparing the measured and predicted wind erosion rates for the training and testing data, respectively. The acceptable levels of the statistical validation criteria were also an indication of the proper performance of the model. Furthermore, the soil erodibility was sensitive respectively to surface crust, very fine sand, and very coarse sand parameters.

An Efficient Machine-Learning Approach for PDF Tabulation in Turbulent Combustion Closure

ABSTRACT Probability density function (PDF) based turbulent combustion modeling is limited by the need to store multi-dimensional PDF tables that can take up large amounts of memory. A significant saving in storage can be achieved by using various machine-learning techniques that represent the thermo-chemical quantities of a PDF table using mathematical functions. These functions can be computationally more expensive than the existing interpolation methods used for thermo-chemical quantities. More importantly, the training time can amount to a considerable portion of the simulation time. In this work, we address these issues by introducing an adaptive training algorithm that relies on multi-layer perception (MLP) neural networks for regression and self-organizing maps (SOMs) for clustering data to tabulate using different networks. The algorithm is designed to address both the multi-dimensionality of the PDF table as well as the computational efficiency of the proposed algorithm. SOM clustering divides the PDF table into several parts based on similarities in data. Each cluster of data is trained using an MLP algorithm on simple network architectures to generate ‘local’ functions for thermo-chemical quantities. The algorithm is validated for the so-called DLR-A turbulent jet diffusion flame using both RANS and LES simulations and the results of the PDF tabulation are compared to the standard linear interpolation method. The comparison yields a very good agreement between the two tabulation techniques and establishes the MLP-SOM approach as a viable method for PDF tabulation.

Artificial Neural Network based computing model for wind speed prediction: A case study of Coimbatore, Tamil Nadu, India

The two main challenges of predicting the wind speed depend on various atmospheric factors and random variables. This paper explores the possibility of developing a wind speed prediction model using different Artificial Neural Networks (ANNs) and Categorical Regression empirical model which could be used to estimate the wind speed in Coimbatore, Tamil Nadu, India using SPSS software. The proposed Neural Network models are tested on real time wind data and enhanced with statistical capabilities. The objective is to predict accurate wind speed and to perform better in terms of minimization of errors using Multi Layer Perception Neural Network (MLPNN), Radial Basis Function Neural Network (RBFNN) and Categorical Regression (CATREG). Results from the paper have shown good agreement between the estimated and measured values of wind speed. According to the result, it can be concluded that ANN model with MLPNN could produce the acceptable prediction of the wind speed for given on wind direction.

Advanced Coherent System For Predicting Cardiac Risks using Data Mining Techniques

Considering health care and medical industry related data there are millions or tons of data which contains numerous hidden information. This information can be mined through which we can make effective decisions in their related industry. There are numerous far advanced methods and techniques in mining and determining the useful decisions using the retrieved useful information. Such an effective system called Coherent cardiac risk prediction system (CCRPS) is developed using neural networks in early detection or prediction of various risk level in cardiac disease. This work employs a multilayer perception neural network with back propagation as the training algorithm. This system aims in predicting the likelihood of patients getting disease related to cardiac such as CHD, a prior heart attack, uncontrolled hypertension, abnormal heart valves, congenital heart disease (heart defects present at birth) and heart muscle disease. The system uses a total of twenty-one medical related parameters such as age, sex, chest pain type, resting blood pressure (in mm Hg on admission to the hospital), serum cholesterol in mg/dl, Smoking, stress etc for prediction purpose. It enables or activated the important knowledge such as how the medical factors related to cardiac disease and patterns and the relationship to be established. Through this system we obtain effective results that have crafted its own diagnostic method or way to predict the risk level measurement of cardiac disease.

Development of soil spectral allocation models considering the effect of soil moisture

Soil spectral allocation or classification is usually conducted on air-dried soils. However, the field soils are not all air-dried, and the change of soil moisture will affect soil reflectance. We introduce a soil allocation model that considers the effect of soil moisture for the purpose of eliminating the effect of soil moisture. The topsoil spectral curves of four typical soils from the Songnen Plain in Northeast China were re-sampled to 10-nm intervals and converted to first-derivative spectral curves and continuum removal curves. The spectral feature parameters were extracted from continuum removal curves in the visible-near infrared (VNIR) range (350–2500 nm), and the range of 430–2400 nm was used to build soil allocation models for reducing the effect of noise. Samples with different soil moisture were mixed into air-dried soils and we calculated the coefficient of variation (CV) of different inputs to assess the effect of soil moisture and to find allocation indices that were not affected by soil moisture. We used allocation indices of Zhang et al. (2018) because of the high accuracy of their DT (Decision Tree) model to allocate mixed-soil samples. We also used allocation indices that were not affected by soil moisture to allocate mixed-soil samples with decision tree (DT), multinomial logistic regression (MLR) and multi-layer perception neural network (MLPNN), and compared the results of the two methods. The results show the following: 1) As SFPs were built with shorter bands, SFP was less sensitive to soil moisture than PCR and PCFD and thus SFP is more suitable to build soil allocation models that consider the effect of soil moisture as input than PCR and PCFD. 2) Differences in soil moisture had little effect on absorption valley shoulders, symmetry and absorption positions, moderate effect on absorption area and depth, and a major effect on the slope of different bands. 3) The effect of soil moisture on continuum removal curves of different soil classes was variable. There was little effect on Arenosols, a moderate effect on Chernozems and Cambisols, and a large effect on Phaeozems. 4) The accuracy of the DT model using allocation indices that were not affected by soil moisture was 91.892% with a Kappa coefficient of 0.888. Our results suggest that it is feasible to build soil spectral allocation models that are not affected by soil moisture, and this improves the universality of soil spectral allocation, especially to field soils, which can be of considerable help in soil classification.

Optimal feature selection for prediction of wind erosion threshold friction velocity using a modified evolution algorithm

Threshold friction velocity (u⁎t) is a very important parameter, which represents wind erosion potential. Because of the difficulty of measuring u⁎t, it would be advantageous if u⁎t could be estimated indirectly from its effecting factors that can be easily measured. The main purpose of this research was to quantify relationships between u⁎t and various topsoil features using inexpensive approaches. To prepare a reliable dataset, we used a portable wind tunnel for measuring u⁎t at a total of 118 observation points in Kerman province, southeast Iran. We developed a non-dominated sorting genetic algorithm II (NSGA-II), specifically designed to operate with artificial neural network (ANN) to select the most determinant properties that influence u⁎t. A permutation of nine input features including surface crust (SC), gravel coverage (GC), very fine sand (VFS), fine sand (FS), very coarse sand (VCS), electrical conductivity (EC), sodium adsorption ratio (SAR), calcium carbonate equivalent (CCE), and mean weight diameter (MWD), was introduced as explanatory variables. We also examined the potential power of using a Multi-Layer Perception (MLP) neural network for prediction of u⁎t changes in response to spatial variation of the selected features. The results of constructed MLP model revealed the ability of the model for u⁎t prediction and showed that the coefficient of determination (R2) values were 0.91 and 0.89 for training and testing data, respectively. Furthermore, acceptable level of the statistical validation criteria verified reliable performance of the MLP model. This research provided a powerful basis for prediction of u⁎t from topsoil features and surface roughness in arid and semi-arid areas of Iran; however, its generic framework could be used to other arid and semi-arid regions with similar challenges.

Combining functional near-infrared spectroscopy and EEG measurements for the diagnosis of attention-deficit hyperactivity disorder

Recently multimodal neuroimaging which combines signals from different brain modalities has started to be considered as a potential to improve the accuracy of diagnosis. The current study aimed to explore a new method for discriminating attention-deficit hyperactivity disorder (ADHD) patients and control group by means of simultaneous measurement of electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS). Twenty-three pre-medicated combined type ADHD children and 21 healthy children were included in the study. Nonlinear brain dynamics of subjects were obtained from EEG signal using Higuchi fractal dimensions and Lempel–Ziv complexity, latency and amplitude values of P3 wave obtained from auditory evoked potentials and frontal cortex hemodynamic responses calculated from fNIRS. Lower complexity values, prolonged P3 latency and reduced P3 amplitude values were found in ADHD children. fNIRS indicated that the control subjects exhibited higher right prefrontal activation than ADHD children. Features are analyzed, looking for the best classification accuracy and finally machine learning techniques, namely Support Vector Machines, Naive Bayes and Multilayer Perception Neural Network, are introduced for EEG signals alone and for combination of fNIRS and EEG signals. Naive Bayes provided the best classification with an accuracy rate of 79.54% and 93.18%, using EEG and EEG-fNIRS systems, respectively. Our findings demonstrate that utilization of information by combining features obtained from fNIRS and EEG improves the classification accuracy. As a conclusion, our method has indicated that EEG-fNIRS multimodal neuroimaging is a promising method for ADHD objective diagnosis.

A hybrid approach using machine learning and genetic algorithm to inverse modeling for single sphere scattering in a Gaussian light sheet

Light scattering has been proven to be an effective tool to characterize and classify particles of different properties. However, inverse modeling to quantitatively retrieve the particle property from light scattering is still a tough task in most applications. In this paper, a hybrid approach using machine learning and genetic algorithm is developed to obtain the geometrical and optical parameters of a sphere from its angular scattering pattern in a light sheet. Scattering patterns related to different parameters are first generated by numerically solving Mie scattering based on angular spectrum theory. Multilayer perception neural network (NN) is then employed to roughly estimate the parameter, while genetic algorithm is adopted to retrieve the precise value. Influences of intensity noise on the inverse modeling are finally examined. Results suggest that the proposed hybrid approach can retrieve the parameters of the sphere from its scattering pattern with high precision in a time-effective manner, which could be widely applied in various scattering-based instruments.

Evaluating windowing-based continuous S-transform with neural network classifier for detecting and classifying power quality disturbances

The aim of this paper is to evaluate the implementation of windowing-based Continuous S-Transform (CST) techniques, namely, one-cycle and half-cycle windowing with Multi-layer Perception (MLP) Neural Network classifier. Both, the techniques and classifier are used to detect and classify the Power Quality Disturbances (PQDs) into one of possible classes, voltage sag, swell and interrupt disturbance signal. For realizing evaluation, we proposed the methodology that include the PQD generation, the signal detection using windowing-based CST, the features extraction from S-contour matrices, PQD classification using MLP classifier. Then, we perform two type of assessments. Firstly, the accuracy assessment of chosen classifier in relation to three different training algorithms. Secondly, the execution time comparison of the training algorithms. Based on assessment results, we outline several recommendations for future work.

Day-Ahead Forecasting of Hourly Photovoltaic Power Based on Robust Multilayer Perception

Photovoltaic (PV) modules convert renewable and sustainable solar energy into electricity. However, the uncertainty of PV power production brings challenges for the grid operation. To facilitate the management and scheduling of PV power plants, forecasting is an essential technique. In this paper, a robust multilayer perception (MLP) neural network was developed for day-ahead forecasting of hourly PV power. A generic MLP is usually trained by minimizing the mean squared loss. The mean squared error is sensitive to a few particularly large errors that can lead to a poor estimator. To tackle the problem, the pseudo-Huber loss function, which combines the best properties of squared loss and absolute loss, was adopted in this paper. The effectiveness and efficiency of the proposed method was verified by benchmarking against a generic MLP network with real PV data. Numerical experiments illustrated that the proposed method performed better than the generic MLP network in terms of root mean squared error (RMSE) and mean absolute error (MAE).

A monarch butterfly optimization-based neural network simulator for prediction of siro-spun yarn tenacity

Yarn tenacity directly affects the winding and knitting efficiency as well as warp and weft breakages during weaving process and therefore, is considered as the most important parameter to be controlled during yarn spinning process. Yarn tenacity is dependent on fiber properties and process parameters. Exploring the relationship between fiber properties, process parameters and yarn tenacity is very important to optimize the selection of raw materials and improve yarn quality. In this study, an efficient monarch butterfly optimization-based neural network simulator called MBONN was developed to predict the tenacity of siro-spun yarns from some process parameters and fiber properties. To this end, an experimental dataset was obtained with fiber fineness, yarn twist factor, yarn linear density and strand spacing as the input variables and yarn tenacity as the output parameter. In the proposed MBONN, a monarch butterfly optimization algorithm is applied as a global search method to evolve weights of a multilayer perception (MLP) neural network. The prediction accuracy of the MBONN was compared with that of a MLP neural network trained with back propagation algorithm, MLP neural network trained with genetic algorithms and linear regression model. The results indicated that the prediction accuracy of the proposed MBONN is statistically superior to that of other models. The effect of fiber fineness, yarn linear density, twist factor and strand spacing on yarn tenacity was investigated using the proposed MBONN. Additionally, the observed trends in variation of yarn tenacity with fiber and process parameters were discussed with reference to the yarn internal structure. It was established that higher migration parameters result in increasing the siro-spun yarn tenacity. It was found that the yarns with higher migration parameters benefit from a more coherent self-locking structure which severely restricts fiber slippage, thereby increasing the yarn tenacity.

The minimum level for soil allocation using topsoil reflectance spectra: Genus or species?

Spectroscopy has been used for rapid determination of soil physicochemical parameters and soil classification or allocation, usually at the great group level of soil genetic classification. Soil genetic classification is still the main system of classification used in China, on which the Chinese Soil Classification System is based, and includes order, suborder, great group, subgroup, genus, species, and subspecies. In this paper, we introduce a soil allocation model that uses topsoil spectral characteristics and determined the minimum level for soil allocation. The topsoil spectra of four typical soils from the Songnen Plain in Northeast China were used to extract spectral feature parameters (SFPs) with clear physical meaning for soil allocation at the great group, genus, and species levels, and principal components of reflectance and first-derivative reflectance were used for allocation to determine the optimal input. Multinomial logistic regression MLR, multi-layer perception neural network, MLPNN, and decision trees (DT) were used to build allocation models. The results show the following: 1) the first and second absorption valleys of the topsoil spectral curves are substantially different between soils at the great group and genus levels but similar at the species level; 2) the topsoil samples can be allocated with SFPs, for which genus is the minimum level within the Chinese Soil Classification System for soil allocation with topsoil hyperspectral reflectance; and 3) the DT model with SFPs as input is the most accurate at the genus level; the accuracy of allocation is 83.3% and the Kappa coefficient, an evaluation index, is 0.8. Our results suggest that soil allocations using topsoil hyperspectral reflectance can be performed accurately at genus level of classification, which can be of considerable help in research on the effects of soil minerals on soil spectral characteristics and in detailed soil mapping.

Cerebral and gaze data fusion for wheelchair navigation enhancement: case of distracted users

SUMMARYThe goal of this paper is to present a new hybrid system based on the fusion of gaze data and Steady State Visual Evoked Potentials (SSVEP) not only to command a powered wheelchair, but also to account for users distraction levels (concentrated or distracted). For this purpose, a multi-layer perception neural network was set up in order to combine relevant gazing and blinking features from gaze sequence and brainwave features from occipital and parietal brain regions. The motivation behind this work is the shortages raised from the individual use of gaze-based and SSVEP-based wheelchair command techniques. The proposed framework is based on three main modules: a gaze module to select command and activate the flashing stimuli. An SSVEP module to validate the selected command. In parallel, a distraction level module estimates the intention of the user by mean of behavioral entropy and validates/inhibits the command accordingly. An experimental protocol was set up and the prototype was tested on five paraplegic subjects and compared with standard SSVEP and gaze-based systems. The results showed that the new framework performed better than conventional gaze-based and SSVEP-based systems. Navigation performance was assessed based on navigation time and obstacles collisions.

A Grey Wolf Optimizer-based neural network coupled with response surface method for modeling the strength of siro-spun yarn in spinning mills

The tenacity of spun yarns is related to many process parameters and fiber properties. Different types of predictive models have been developed to predict the spun yarns tensile strength; however, no investigation has yet been carried out on prediction of siro-spun yarns tensile strength. This is due to the fact that the relationship between yarn strength and fiber properties and process parameters is essentially non-linear and therefore the prediction of yarn tenacity is a highly complex issue. This paper proposes a Grey Wolf Optimizer (GWO) based Neural Network Simulator called “GWNN” for prediction of siro-spun yarn tensile strength. In the proposed GWNN, a GWO algorithm is applied as a global search method to determine weights of a Multi-Layer Perception (MLP). Additionally, a new Response Surface Methodology (RSM) is proposed to determine appropriate level of fiber and yarn linear densities and processing parameters on yarn tensile strength. The proposed RSM uses the GWNN model as a non-linear response surface simulator and has higher accuracy than the classical RSM. The prediction accuracy of the GWNN was compared with that of a MLP neural network trained with Back-Propagation (BP) algorithm and a Multiple Linear Regression model as well as three evolutionary-based neural networks. It was found that the proposed GWNN enjoys higher accuracy as compared with other models. Additionally, the observed trends in variation of yarns tensile strength with input variables were discussed with reference to inner structure of yarns.

Thrust and torque force analysis in the drilling of aramid fibre-reinforced composite laminates using RSM and MLPNN-GA

Aramid Fibre Reinforced Plastic composites are difficult to be drilled due to anisotropic material properties. Currently, soft computing techniques are used as alternatives to conventional mathematical models, which is robust and can deal with inaccuracy and uncertainty. In this paper, drilling of Aramid Fibre Reinforced Plastics (AFRPs) was carried out using Taguchi L54 experimental layout. Drilling tool used in this experiment was solid carbide. The purpose of this study was to find optimum combination of drilling parameters to obtain minimum thrust and torque force to reduce the delamination. Also, this paper proposed a prediction model of Multilayer Perception Neural Network optimized by Genetic Algorithm (MLPNN-GA). Moreover, RSM technique was used to evaluate the influence of process parameters (spindle speed, feed rate, drill point angle and drill diameter on thrust force and torque. The prediction capability of both RSM and MLPNN-GA was compared with Response optimizer for thrust force and torque. The investigation demonstrated that drill point angle is the primary factor affecting thrust force and drill diameter influences the torque force on the drill bit. Overall, this study recommends the use of high speed and low feed combination and drill point angles of 90°–118° to reduce the delamination of the materials in the drilling of AFRP composites.

Automated Chest X-Ray Screening: Can Lung Region Symmetry Help Detect Pulmonary Abnormalities?

Our primary motivator is the need for screening HIV+ populations in resource-constrained regions for exposure to Tuberculosis, using posteroanterior chest radiographs (CXRs). The proposed method is motivated by the observation that radiological examinations routinely conduct bilateral comparisons of the lung field. In addition, the abnormal CXRs tend to exhibit changes in the lung shape, size, and content (textures), and in overall, reflection symmetry between them. We analyze the lung region symmetry using multi-scale shape features, and edge plus texture features. Shape features exploit local and global representation of the lung regions, while edge and texture features take internal content, including spatial arrangements of the structures. For classification, we have performed voting-based combination of three different classifiers: Bayesian network, multilayer perception neural networks, and random forest. We have used three CXR benchmark collections made available by the U.S. National Library of Medicine and the National Institute of Tuberculosis and Respiratory Diseases, India, and have achieved a maximum abnormality detection accuracy (ACC) of 91.00% and area under the ROC curve (AUC) of 0.96. The proposed method outperforms the previously reported methods by more than 5% in ACC and 3% in AUC.

Research on electronic nose system based on continuous wide spectral gas sensing

Light absorption gas sensing technology has characteristics of massive parallelism, cross-sensitivity and extensive responsiveness which make it suitable for the sensing task of an electronic nose (e-nose). Therefore, an innovational e-nose system based on continuous wide spectral (CWS) gas sensing was proposed in this paper, named as optical e-nose. The mathematical model of the system was established and corresponding experimental platform was built. Then four kinds of air contaminations (NO2, SO2, the mixture of SO2 and NO2 (SO2&NO2), C6H6) were chosen to test the platform and corresponding spectra were collected to verify the feasibility of the proposed system. In addition, typical classifiers like correlation coefficient (CC), Euclidean distance to centroids (EDC), k-nearest neighbor (KNN), multilayer perception neural network (MLP), support vector machine (SVM) and least squares support vector machine (LSSVM) were chosen for classification analysis. Experimental results show that the mean classification accuracy of experimental data is >90% which confirmed the effectiveness of the optical e-nose.

Prediction of thermal conductivities of polyacrylonitrile electrospun nanocomposite fibers using artificial neural network and prey predator algorithm

In this study, artificial neural networks were employed to predict thermal conductivity of polyacrylonitrile (PAN) electrospun nanocomposite fibers embedded with multiwalled carbon nanotubes (MWCNTs) and Nickel Zinc ferrites [(Ni0.6Zn0.4) Fe2O4]. The prey predator algorithm was used to train the neural networks to find the best models. This is the first paper on the application of multilayer perception neural network (MLPNN) with the prey predator algorithm for the prediction of thermal conductivity of PAN Electrospun nanocomposites. The method of nonlinear regression was used to minimize the error distribution between the experimental data and the predicted results. Both MWCNTs and Nickel Zinc ferrites were used in different weight proportions. The predicted ANN responses were analyzed statistically using z-test and error functions for both nanoinclusions. The predicted ANN responses for PAN Electrospun nanocomposite fibers were compared with the experimental data and were found in good agreement.

Evaluation and prediction of blast-induced peak particle velocity using artificial neural network: A case study

Over the past few decades, inducing of ground vibrations from blasting may cause severe damage to surrounding structures, plants, and human beings in the mining industry. Therefore, it is essential to monitor and predict the ambiguous vibration levels and take measures to reduce their hazardous effect. In this study, to evaluate and predict the ambiguous ground vibrations, an application of artificial neural network technique was used. A three-layer, feed-forward back-propagation multilayer perception neural network having six input parameters, the distance from blast face, maximum charge per delay, spacing, burden, hole depth and a number of holes, and one output: peak particle velocity, was used and trained with the Levenberg–Marquardt algorithm using 25 experimental and blast event records from the iron ore mine A, India. To determine the efficiency and accuracy of the developed artificial neural network model, seven conventional predictor models proposed by the US Bureau of Mines, Ambraseys–Hendron, Langefors–Kihlstrom, general predictor, Ghosh–Daemen predictor, cardiac magnetic resonance imaging (CMRI) predictor, Bureau of Indian Standards, as well as multiple linear regression, were applied to establish a relation between peak particle velocity and its influencing parameters. The obtained results reveal that the proposed artificial neural network model can estimate ground vibrations more accurately as compared with the various conventional predictor models available. Coefficient of determination (R2) and root mean square error indices were obtained as 0.9971 and 0.08133 for artificial neural network model, respectively.

Comparing Two Methods of Neural Networks to Evaluate Dead Oil Viscosity

Reservoir characterization and asset management require comprehensive information about formation fluids. In fact, it is not possible to find accurate solutions to many petroleum engineering problems without having accurate pressure-volume-temperature (PVT) data. Traditionally, fluid information has been obtained by capturing samples and then by measuring the PVT properties in a laboratory. In recent years, neural network has been applied to a large number of petroleum engineering problems. In this paper, a multi-layer perception neural network and radial basis function network (both optimized by a genetic algorithm) were used to evaluate the dead oil viscosity of crude oil, and it was found out that the estimated dead oil viscosity by the multi-layer perception neural network was more accurate than the one obtained by radial basis function network.