Feed-forward Back-propagation Neural Network Research Articles

Slime Mould Algorithm Training Neural Network in Automatic Voltage Regulator

The research is proposed a new method of artificial intelligence (AI) to control automatic voltage regulators. A neural network has improved using a metaheuristic method, namely the slime mould algorithm (SMA). SMA has an algorithm based on the mode of slime mold in nature. SMA has characteristics that use adaptive weights to simulate the process to generate feedback from the movement of bio-oscillator-based slime molds in foraging, exploring, and exploiting areas. The performance of the proposed method is focused on speed and rotor angle. To know the competence and potency of the proposed method, a comparison with feed-forward backpropagation neural networks (FFBNN), cascade-forward backpropagation neural networks (CFBNN), Elman-recurrent neural networks (E-RNN), focused time delay neural network (FTDNN), and Distributed Time Delay Neural Network (DTDNN) method are applied. It can be concluded that the proposed method has the best ability. The Proposed method has ability to reduce the overshoot speed with an average value of 0.78 % and the overshoot rotor angle with an average value of 2.134 %.

A comprehensive investigation and artificial neural network modeling of shape stabilized composite phase change material for solar thermal energy storage

Shape stabilized composite phase change materials (SSCPCM) are an attractive option for solar thermal energy storage applications. The objective of the present study is to understand and quantify the shape stability of a CPCM and its kinetics behavior. A Feed-Forward Backpropagation Neural Network (FFBPNN) algorithm is developed for the prediction of phase change temperature and heat flow of composite PCMs. Myristyl alcohol mixed with expanded graphite at various loadings (5%, 10%, and 15%) are used to study the shape stability of CPCM. The characterization of PCM and CPCM is done using Fourier-transform analysis, X-ray diffraction analysis, Scanning electron microscope, and Brunauer-Emmett-Teller (BET) analysis, Differential scanning calorimetry, and Thermogravimetric analysis. The thermophysical properties of PCM and CPCM are investigated using the C-Therm instrument. DSC experiments at different heating rates are performed to estimate the activation energy of the PCM and CPCM. The SSCPCM showed 11.82%, 11.70% higher activation energy than PCM as determined by Kissinger and Ozawa models. The activation energy of CPCM increases with the loading of expanded graphite. The charging and discharging time of SSCPCM is 29.8%, 23.3% lower than the PCM, respectively. The thermal conductivity of SSCPCM is 17.6 times higher than the PCM. Therefore, SSCPCM is an attractive solution to replace the conventional available PCMs, owing to no leakage problem and a considerable improvement in thermal conductivity. Artificial Neural Network (3–18–2) topology effectively predicts the phase change temperature and heat flow of shape stabilized composite PCMs with a coefficient of determination greater than 0.989, which is helpful in the design and selection of PCM for latent heat thermal energy storage systems.

Designing of High Voltage Cable Bonding with Intelligence Algorithms to Avoid Cable Insulation Faults and Electroshock in High Voltage Lines

The insulation fault is a major problem in high voltage cable lines. The major factors in the insulation faults are harmonic currents and the metal sheath voltage (MV) that occur on the metal sheath of cable. MV and harmonic distortion should be minimized to prevent insulation faults. Thus, sectional solid bonding with different grounding resistance (SSBr) method is developed as a new bonding method for minimizations of harmonic current and MV. Also, SSBr should be optimized by optimized according to minimum MV and harmonic distortion rate of high voltage cable. Inertia weighted particle swarm optimization (iPSO), particle swarm optimization (PSO), genetic algorithm (GA) and differential evolution algorithm (DEA) are used for optimization of SSBr, and three groups the prediction methods are used separately as objective function of the optimization methods to determine minimum MV and harmonic distortion. These groups are neural networks, hybrid neural networks and regression methods. Hybrid neural network with inertia weighted particle swarm optimization (H-iPSO), linear regression and feedforward backpropagation neural network are selected from their groups according to training errors. Solid bonding method is widely used for bonding of high voltage cable, and solid bonding is simulated in this study. When solid bonding is used for bonding of cable, maximum harmonic distortion rate is measured as 8.15 %, and maximum MV is measured as 1086 V. When H-iPSO is used as prediction method, and PSO is used as optimization method, maximum harmonic distortion rate is measured as 5,28 %, and maximum MV is measured as 57 V. Namely, both insulation fault and electroshock can be prevented by the optimized SSBr method.

Tuning Rules for Fractional Order PID Controller Using Data Analytics

Flexibility and robust performance have made the FOPID (Fractional Order PID) controllers a better choice than PID (Proportional, Integral, Derivative) controllers. But the number of tuning parameters decreases the usage of FOPID controllers. Using synthetic data in available FOPID tuners leads to abnormal controller performances limiting their applicability. Hence, a new tuning methodology involving real-time data and overcomes the drawbacks of mathematical modeling is the need of the hour. This paper proposes a novel FOPID controller tuning methodology using machine learning algorithms. Feed Forward Back Propagation Neural Network (FFBPNN), Multi Least Squares Support Vector Regression (MLSSVR) chosen to design Machine Learning based Optimal Tuner (MLOT) can handle the interdependency between the controller parameters and multiple outputs for multiple inputs.The proposed tuner finds application in the control of power and energy systems. It can accomplish tracking, disturbance-rejection, and robustness controller performances, thus making FOPID controller design easier and accurate. Comparisons with existing FOPID tuning rules show better controller performances and easy tuning. Thus, this paper addresses a unique, real-time, model-free, easily tunable FOPID tuning methodology satisfying plant requirements.

An anti-phishing model based on similarity measurement

Phishing has represented a more noteworthy danger to clients. In the current work, the authors attempted to build up a powerful anti-phishing technique based on hybrid similarity approach combining Cosine and Soft Cosine similarity that measures the resemblance between user query and database. The proposed similarity hybrid is also evaluated against another similarity hybrid comprising of Cosine and Jaccard similarity measure so as to validate the proposed work. Both hybrid similarities are separately fed to validation layer of feed forward back propagation neural network (FFBPNN) to predict phishing and legitimate websites. The performance of the proposed work is evaluated against data set comprising of 3,000 sample files in terms of positive predictive value (PPV), true positive rate (TPR) and F-measure. The comparative analysis demonstrated that the anti-phishing model using proposed similarity hybrid outperformed the cosine and Jaccard similarity hybrid with 0.233%, 0.2833% and 0.258% higher PPV, TPR and F-measure, respectively.

Coverage Properties of a Neural Network Estimator of Finite Population Total in High‐Dimensional Space

The problem in nonparametric estimation of finite population total particularly when dealing with high‐dimensional datasets is addressed in this paper. The coverage properties of a robust finite population total estimator based on a feedforward backpropagation neural network developed with the help of a superpopulation model are computed, and a comparison with existing model‐based estimators that can handle high‐dimensional datasets is conducted to evaluate the estimator’s performance using simulated datasets. The results presented in this paper show good performance in terms of bias, MSE, and mean absolute error for the feedforward backpropagation neural network estimator as compared to other identified existing estimators of finite population total in high‐dimensional datasets. In this regard, the paper recommends the use of the proposed estimator in estimating population parameters such as population total in the presence of high‐dimensional datasets.

Technical Note: A Comparison of Supervised Machine Learning Algorithms for Predicting Subfield Yield Variability of Maize Grain

Abstract. Precision farming is about managing within-field spatial variability by applying appropriate inputs with the right amount, at the right place, and at the right time. This typically leads to the delineation of site-specific management zones that represent different yield potentials. However, a focus on spatiotemporal interactions is generally lacking. This technical note explores the viability of predicting spatial yields within fields, in the small-field arable production commonly practiced in New Zealand, using supervised machine learning algorithms. The methods used are multiple linear regression (MLR), feed-forward backpropagation neural network (FFNN), classification and regression tree (CART), random forest (RF), XGBoost, and Cubist, using predictors compiled from a range of spatiotemporal factors, including soil electrical conductivity (EC), soil organic matter (OM), elevation, rainfall, growing degree days, and solar radiation. Despite poor results (R2 = 0.06 to 0.50) for predicting the spatial yield of each year, the RF, XGBoost, and Cubist models demonstrated greater capability for modeling spatiotemporal interactions than the previously tested FFNN and MLR. The inclusion of consistently calibrated yield data for additional years and more related variables (e.g., soil moisture and canopy cover) could improve the models. The results of these modeling analyses could lead to the delineation of dynamic yield management zones for improving the precision of mid-season fertilizer prescriptions to improve yield. Keywords: GIS, Machine learning, Precision farming, Spatiotemporal.

Estimation of Finite Population Totals in High Dimensional Spaces

In this paper, the problem of Nonparametric Estimation of Finite Population Totals in high dimensional datasets is considered. A robust estimator of the Finite Population Total based on Feedforward Backpropagation Neural Network is derived with the aid of a Super-Population Model. This current study is motivated by the fact that Local Polynomials and Kernel methods have in preceding related studies, been shown to provide good estimators for Finite Population Totals but in low dimensions. Even in these situations however, bias at boundary points presents a big challenge when using these estimators in estimating Finite Population parameters. The challenge worsens as the dimension of regressors increase. This is because as the dimension of the Regressor Vectors grows, the Sparseness of the Regressors’ values in the design space becomes unfeasible, resulting in a decrease in the fastest achievable rates of convergence of the Regression Function Estimators towards the target curve, rendering Kernel Methods and Local Polynomials ineffective to address these challenges. This study considers the technique of Artificial Neural Networks which yields robust estimators in high dimensions and reduces the estimation bias with marginal increase in variance. This is due to its Multi-Layer Structure, which can approximate a wide range of functions to any required level of precision. The estimator’s properties are developed, and a comparison with existing estimators was conducted to evaluate the estimator’s performance using real data sets acquired from the United Nations Development Programme 2020. The estimation approach performs well in an example using data from a United Nations Development Programme 2020 on the study of Human Development Index against other factors. The theoretical and practical results imply that the Neural Network estimator is highly recommended for survey sampling estimation of the finite population total.

Inversion of 2D cross-hole electrical resistivity tomography data using artificial neural network.

Geophysical inversion is often ill-posed because of its nonlinearity and the ordinary measured data of measured data. To deal with these problems, an artificial neural network (ANN) has been introduced with the capability of a nonlinear and complex problem for geophysical inversion. This study aims to invert 2D cross-hole electrical resistivity tomography data using a feedforward back-propagation neural network (FBNN) approach. To generate the synthetic data to train the model, eighteen forward models (100 to 600 Ω.m homogeneous medium and three different locations of 10 Ω.m of the grouted bulb) with a dipole-dipole array configuration were adopted. The effect of the hyperparameter on the performance of the proposed FBNN model was examined. Various datasets from the laboratory testing result were also tested using the suggested FBNN model and then the error between the actual and predicted area in each model was determined. The results show that our suggested FBNN model, with the trainrp training function, 4 hidden layers, 75 neurons in each hidden layer, 0.8 learning rate, 1 of momentum coefficient, and 54,000 training data points, has higher performance and better accuracy than other models. It was found that the error value of the FBNN model was about 15% to 18% lower compared to the conventional inversion model.

Three-Phase Feeder Load Balancing Based Optimized Neural Network Using Smart Meters

The electricity distribution system is the coupling point between the utility and the end-user. Typically, these systems have unbalanced feeders due to the variety of customers’ behaviors. Some significant problems occur; the unbalanced loads increase the operational cost and system investment. In radial distribution systems, swapping loads between the three phases is the most effective method for phase balancing. It is performed manually and subjected to load flow equations, capacity, and voltage constraints. Recently, due to smart grids and automated networks, dynamic phase balancing received more attention, thus swapping the loads between the three phases automatically when unbalance exceeds permissible limits by using a remote-controlled phase switch selector/controller. Automatic feeder reconfiguration and phase balancing eliminates the service interruption, enhances energy restoration, and minimize losses. In this paper, a case study from the Irbid district electricity company (IDECO) is presented. Optimal reconfiguration of phase balancing using three techniques: feed-forward back-propagation neural network (FFBPNN), radial basis function neural network (RBFNN), and a hybrid are proposed to control the switching sequence for each connected load. The comparison shows that the hybrid technique yields the best performance. This work is simulated using MATLAB and C programming language.

Spatial quantification method of grassland utilization intensity on the Qinghai-Tibetan Plateau: A case study on the Selinco basin

Existing methods for spatial quantification of grassland utilization intensity cannot meet the demand for accurate detection of the spatial distribution of grassland utilization intensity in the Qinghai-Tibetan Plateau with high spatial resolution. In this paper, a method based on remote-sensing observations and simulations of grassland growth dynamics is proposed. The grassland enhanced vegetation index (EVI) time-series curve during the growing season characterizes the growth of grassland in the corresponding pixel; The deviation between the observed and potential EVI curves indicates the disturbance on grassland growth imposed by human activities, and it can characterize the grassland utilization intensity during the growing season. Based on the main idea described above, absolute and relative disturbances are calculated and used as quantitative indicators of grassland utilization intensity defined from different perspectives. Livestock amount at the pixel scale is obtained by pixel-by-pixel calculations based on the function relationship at the township scale between absolute disturbance and livestock density, which is specific quantitative indicator that considers the mode of grassland utilization. In simulating the potential EVI of grassland, the lag and accumulation effects of meteorological factors are investigated at the daily scale using a multi-objective genetic algorithm. Further, the nonlinear functions between multiple environmental factors (e.g., grassland type, topography, soil, meteorology) and the grassland EVI are established using an error back-propagation feedforward artificial neural network (ANN-BP) with parameter optimization. Finally, the potential EVIs of all grassland pixels are simulated on the basis of this model. The method is applied to the Selinco basin on the Qinghai-Tibetan Plateau and validated by examining the spatial consistency of the results with township-scale livestock density and grazing pressure. The final results indicate that the proposed method can accurately detect the spatial distribution of grassland utilization intensity which is appliable in the similar regions.

VOC removal in rotating packed bed: ANN model vs empirical model

The discharged volatile organic compounds (VOCs) have aroused more and more attention because of increasing serious air pollution. Accurate prediction of the emission is very important for the industry. Therefore, two kinds of model were developed to predict the VOCs removal efficiency in the RPB, including empirical model and artificial neural network (ANN) models. The empirical model was mainly including gas and liquid mass transfer, effective contact area and liquid holdup. The ANN models were including Cascade-forward back propagation neural network, Feed-forward distributed time delay neural network, Feed-forward back propagation neural network and Elman-forward back propagation neural network. The input parameters were dimensionless numbers, such as high gravity factor, liquid Reynolds number, gas Reynolds number and dimensionless Henry coefficient. The output parameter was removal efficiency. The outlet concentrations of different VOCs predicted by the ANN model were much better than those of the empirical model. And the disadvantages and the advantages were also analyzed. The effects of high gravity factor, gas flow rate and liquid flow rate on the removal efficiency in the RPB were simulated by ANN model with high precision. And then, operation conditions could be timely adjusted to meet the emission standards.

Modeling of neural networks for weather forecasting in debre markos, Ethiopia

ABSTRACT The use of artificial neural networks (ANNs) to make daily weather predictions based on historical data is proposed in this study. The proposed models are designed using four learning algorithms: feedforward backpropagation, cascade forward backpropagation, feedforward distributed time delay, and layer recurrence. Moreover, the four learning algorithms use TRAINLM, TRAINRP, TRAINSCG, TRAINGDX, and TRAINOSS training functions independently. Weather data of 5500 days were used for training, validating, and testing using the designed networks. By measuring the atmospheric minimum and maximum temperatures, relative humidity, precipitation, sunshine hours, and wind speed on each day, the weather data of the next day, that is, the maximum temperature of the atmosphere and sunshine hours, were forecast. The network architecture is composed of 1 hidden layer and 12 neurons. The feedforward backpropagation neural network achieved a better result than the other neural networks. The numerical predictions of the feedforward backpropagation neural network were nearly equal to the actual values, and its forecast was responsible for a mean square error of less than 2.11 for the overall training. During the training, validation and testing phase the recorded regression were 0.99109, 0.98298 and 0.99184, respectively. The aggregated regression value considering training, validation, and testing was 0.98955.

Damage detection in RC beam utilizing feed-forward backpropagation neural network technique

Damage detection in RC beam utilizing feed-forward backpropagation neural network technique

Modelling the hardened properties of steel fiber reinforced concrete using ANN

The hardened properties of concrete are one of the most important parameters when considering the sustainability criteria of buildings. Numerous previous studies have established how the Artificial Neural Network is used as an effective statistical data modelling tool. In this study, the prediction of Elastic modulus and Compressive Strength of steel fibre reinforced concrete have been illustrated with a feed-forward backpropagation neural network structure. 158 datasets are used for modelling of elastic modulus and 140 datasets is used for modelling of Compressive Strength. The developed ANN models were able to predict the data within the range of input parameters considered and were having regression coefficients values of 0.96 and 0.97 respectively. A comparison of actual and predicted values has been performed and the results indicated that the ANN performed better in terms of prediction with minimum deviation from original data.

Predictive model for progressive salinization in a coastal aquifer using artificial intelligence and hydrogeochemical techniques: a case study of the Nile Delta aquifer, Egypt.

To monitor groundwater salinization due to seawater intrusion (SWI) in the aquifer of the eastern Nile Delta, Egypt, we developed a predictive regression model based on an innovative approach using SWI indicators and artificial intelligence (AI) methodologies. Hydrogeological and hydrogeochemical data of the groundwater wells in three periods (1996, 2007, and 2018) were used as input data for the AI methods. All the studied indicators were enrolled in feature extraction process where the most significant inputs were determined, including the studied year, the distance from the shoreline, the aquifer type, and the hydraulic head. These inputs were used to build four basic AI models to get the optimal prediction results of the used indicators (the base exchange index (BEX), the groundwater quality index for seawater intrusion (GQISWI), and water quality). The machine learning models utilized in this study are logistic regression, Gaussian process regression, feedforward backpropagation neural networks (FFBPN), and deep learning-based long-short-term memory. The FFBPN model achieved higher evaluation results than other models in terms of root mean square error (RMSE) and R2 values in the testing phase, with R2 values of 0.9667, 0.9316, and 0.9259 for BEX, GQISWI, and water quality, respectively. Accordingly, the FFBPN was used to build a predictive model for electrical conductivity for the years 2020 and 2030. Reasonable results were attained despite the imbalanced nature of the dataset for different times and sample sizes. The results show that the 1000 μS/cm boundary is expected to move inland ~9.5 km (eastern part) to ~10 km (western part) to ~12.4 km (central part) between 2018 and 2030. This encroachment would be hazardous to water resources and agriculture unless action plans are taken.

Rapid and simple spectrophotometric method using feedforward backpropagation and radial basis function neural networks for the simultaneous determination of amoxicillin and clavulanic acid in commercial tablet and human blood serum

In this study, spectrophotometry method was combined with artificial intelligence approaches, including feedforward backpropagation neural network (FFBP-NN) and radial basis function neural network (RBF-NN) for the simultaneous estimation of amoxicillin (AMX) and clavulanic acid (CLV) in binary mixtures, pharmaceutical tablet, and blood serum. Levenberg–Marquardt (LM) and Resilient Backpropagation (RP) were applied as training algorithms of FFBP-NN. Among with different layers (2 and 4) and neurons (2, 4, 6, and 8) related to the LM algorithm, layers 4 and 2 with 8 neurons revealed the lowest mean square error (MSE) equal to 4.72 × 10−29 and 4.32 × 10−29 for AMX and CLV, respectively. In the RP algorithm, layers 4 and 2 with 8 neurons indicated minimum MSE of 1.63 and 0.605 for AMX and CLV, respectively. In the RBF-NN, MSE equal to 6.47 × 10−26 and 1.83 × 10−26 was obtained for AMX and CLV, respectively. Analysis of variance (ANOVA) test at the 95% confidence level was used to examine the significant differences between the methods. Also, the determination of the AMX and CLV was simultaneously investigated in blood serum sample as a complex matrix. Simple, fast, and low-cost are the benefits of this spectrophotometry method, which does not require separation step for the simultaneous estimation of components in pharmaceutical formulations.

Removal of reactive orange 16 with nZVI-activated carbon/Ni: optimization by Box-Behnken design and performance prediction using artificial neural networks

PurposeThe purpose of this paper is to analyze nano zero-valent iron (nZVI)-activated carbon/Nickel (nZVI-AC/Ni) by a novel method. The synthesized adsorbent was used to degrade reactive orange 16 (RO 16) azo dye.Design/methodology/approachThe optimum conditions for the highest removal of RO 16 dye were determined. Characterization of nZVI-AC/Ni was done by scanning electron microscopy, Fourier-transform infrared spectroscopy, X-ray diffraction and energy-dispersive X-ray spectroscopy. The nZVI-AC/Ni were used for the removal of dye RO 16 and the parameters affecting were discussed such as pH, adsorbent dosage, contact time and concentration of dye. To investigate the variables and interaction between them, an analysis of variance test was performed.FindingsThe characterization results show that the synthesis of nZVI-AC/Ni caused no aggregation of nanoparticles. The maximum dye removal efficiency of 99.45% occurred at pH 4, the adsorbent dosage = 0.1 gL-1 and the dye concentration of 10 mgL-1. Among various algorithms of feed-forward backpropagation neural network, Levenberg–Marquardt with mean square error (MSE) = 9.86 × 10–22 in layer = 5 and the number of neurons = 9 was selected as the best algorithm. On the other hand, the MSE of the radial basis function model was 0.2159 indicating the good ability of the model to predict the percentage of dye removal.Originality/valueThere are two main innovations. One is that the novel nZVI-AC/Ni was prepared successfully. The other is that the optimized conditions were obtained for the removal of RO 16 dye from an aqueous solution. Furthermore, to the best of the knowledge, no study has ever investigated the removal of RO 16 by nZVI-AC/Ni produced.

Detection of source locations in RC columns using machine learning with acoustic emission data

Detecting and monitoring the damage locations of in-service reinforced concrete (RC) structures is difficult as they exhibit more complex geometries and consist of composite materials. In this study, a new method for detecting the source location of acoustic emission (AE) signals in RC structures using machine learning was introduced and tested on an in-service RC column. An impulse response test was conducted using an AE testing system to gather data regarding the source locations. Meanwhile, the source locations on the test column were generated through manual hits. The corresponding AE data were collected using a AE data acquisition system. The significant AE parameters were determined using neighborhood component analysis for feature selection to obtain a dataset for machine learning. Further, a support vector machine classifier was adopted to classify the source locations of the columns for three sensor groups. Furthermore, the source locations were predicted using feedforward backpropagation neural networks. This study validated the detectability of AE signals in RC columns and proposed a novel detection method, which can be employed to perform continuous nondestructive tests on in-service RC columns.

Gait phase detection by using a portable system and artificial neural network

Gait phases are important to evaluate the walking function and to identify the characteristics of pathological gaits. However, it is difficult to differentiate gait phases outside gait laboratories, thus, this study aimed to develop a method to detect 8 gait sub-phases using a wearable multiple sensor system and artificial neural network (ANN). Motion sensors were used to acquire the acceleration of lower limbs, and force sensitive resistors were used to detect contact state and force between the foot and the ground. Walking was recorded using a high-speed camera. Two feed forward back-propagation (BP) neural networks were developed. The resilient BP algorithm was used to train ANN. A total of 66 volunteers participated in this study. For the stance and swing phase detection, simulation of the training data showed an accuracy of 98.0 ​%. The data from the test set showed a recognition accuracy of 97.75 ​%. Because the ending point of the last phase ‘Terminal Swing’ is always 100 ​% GC, we only listed seven phases. The prediction accuracy of seven phases were: 35.9 ​%, 63.8 ​%, 93.6 ​%, 94.9 ​%, 94.8 ​%, 97.9 ​% and 98 ​% using the limb acceleration data only. The average accuracy for seven phases were 68 ​%, 91.3 ​%, 97.8 ​%, 98.9 ​%, 98.8 ​%, 99.1 ​%, and 99.5 ​% using the limb acceleration and foot pressure data for fast, normal, and slow gait speeds. This study provides a new method for eight gait sub-phases detection with high accuracy combining a wearable system and ANN, which may make gait phase analysis possible under free-living conditions.