Backpropagation Training Algorithm Research Articles

Assessment of strength characteristics for experimental based workable self compacting concrete using artificial neural network

Self compacting concrete (SCC) is the concrete which can flow through reinforcement without the requirement of vibration. The proportioning of materials for workable SCC requires clear understanding of material properties and their behavior with one on one to apply it in construction site. Hence, an extensive experimentation is conducted based on empirical way and produced 123 workable SCC mixtures resulting for measuring the strength characteristics such as compressive strength (fc) and splitting tensile strength (fst). The workable SCC mixture is considered for strength characteristics only after passing the requirements of workability tests of SCC namely slump flow test, L-box test, V-funnel test and sieve segregation resistance test. The 123 workable SCC mixtures experimentally obtained fc ranging from 26.6 N/mm2 to 62.1 N/mm2 and fst ranging from 2.9 N/mm2 to 6.9 N/mm2. In this study, artificial neural network (ANN) had been used which was speedy and efficient for predicting output strength characteristics based on extensive experimental data investigated in laboratory. ANN is data modeling tool where relationship between inputs and output can be found out. In this work, after investigating all parameters for producing SCC, only important inputs considered in ANN model. Marquardt back propagation training algorithm had been used with inputs containing cement content, water cement ratio, type of mineral admixture, percentage of mineral admixture and type of chemical admixture. In this study, satisfactory mean square error and correlation coefficient had been obtained during training, validation and testing of ANN for predicting two outputs fc and fst separately.

Thailand port throughput prediction via particle swarm optimization based neural network

Shipping volume in Thailand have significantly increased in last four years. It is important to pay attention to the trend of Thailand port throughput and use as the guideline to prepare for the needs of supporting facilities, infrastructures, financial and human resources. An effective forecasting technique called particle swarm based neural network (PSONN) is developed to estimate Thailand port throughput in this work. The prediction results from PSONN and classical backpropagation training algorithm, backpropagation neural networks (BPNN) were compared. The results shown that PSONN provides more accurate results than BPNN when apply to predict port throughput of Thailand. The mean squared error obtained from PSONN are about 10 times lower than that of BPNN. This confirms that neural network based on PSO training algorithm has better performance and better ability to escape local optimum than that of BPNN.

Simultaneous feature weighting and parameter determination of Neural Networks using Ant Lion Optimization for the classification of breast cancer

In this paper, feature weighting is used to develop an effective computer-aided diagnosis system for breast cancer. Feature weighting is employed because it boosts the classification performance more as compared to feature subset selection. Specifically, a wrapper method utilizing the Ant Lion Optimization algorithm is presented that searches for best feature weights and parametric values of Multilayer Neural Network simultaneously. The selection of hidden neurons and backpropagation training algorithms are used as parameters of neural networks. The performance of the proposed approach is evaluated on three breast cancer datasets. The data is initially normalized using tanh method to remove the effects of dominant features and outliers. The results show that the proposed wrapper method has a better ability to attain higher accuracy as compared to the existing techniques. The obtained high classification performance validates the work which has the potential for becoming an alternative to the other well-known techniques.

Dataset from estimation of gasification system efficiency using artificial neural network technique

Abstract The dataset contained in this article supplements the data information contained in the performance evaluation of gasification system efficiency using artificial neural network technique. The dataset was obtained from investigating efficiency of a gasification system using 315 fuel samples containing biomass, coal, and coal-biomass blends via the application of Levenberg–Marquardt (LM) back-propagation and Bayesian Regularization (BR) training algorithms in Artificial Neural Networks (ANN) domain. Input Variables Representation Technique-by-Visual Inspection method (IVRT-VIM) and Output Variables Representation Technique –by-Visual Inspection Method (OVRT-VIM) were developed as well to improve results from the ANN models. Sensitivity analysis on the prediction of the product gas compositions (CO, CO2, H2, and CH4) was carried out using partial derivatives (PaD) method. There was a considerable decrease in the MSE and an increase in the R2 of around 93 – 96% and 94 – 96% using 10 & 23 number of neurons in the hidden layer and IVRT-VIM technique, respectively. The L -M algorithm offered a better result than the BR algorithm, and with Sulphur as the most important input variable that affected the outputs.

Prediction of BLEVE mechanical energy by implementation of artificial neural network

In the event of a BLEVE, the overpressure wave can cause important effects over a certain area. Several thermodynamic assumptions have been proposed as the basis for developing methodologies to predict both the mechanical energy associated to such a wave and the peak overpressure. According to a recent comparative analysis, methods based on real gas behavior and adiabatic irreversible expansion assumptions can give a good estimation of this energy. In this communication, the Artificial Neural Network (ANN) approach has been implemented to predict the BLEVE mechanical energy for the case of propane and butane. Temperature and vessel filling degree at failure have been considered as input parameters (plus vessel volume), and the BLEVE blast energy has been estimated as output data by the ANN model. A Bayesian Regularization algorithm was chosen as the three-layer backpropagation training algorithm. Based on the neurons optimization process, the number of neurons at the hidden layer was five in the case of propane and four in the case of butane. The transfer function applied in this layer was a sigmoid, because it had an easy and straightforward differentiation for using in the backpropagation algorithm. For the output layer, the number of neurons had to be one in both cases, and the transfer function was purelin (linear). The model performance has been compared with experimental values, proving that the mechanical energy of a BLEVE explosion can be adequately predicted with the Artificial Neural Network approach.

Performance comparison of feedforward neural network training algorithms in modelling microwave pyrolysis of oil palm fibre for hydrogen and biochar production

AbstractMicrowave pyrolysis of oil palm fibre (OPF) was conducted to study the effect of microwave power, temperature, and nitrogen (N2) flow rate on the production of hydrogen and biochar. The effect of microwave power ranging from 400 to 900 W, reaction temperatures ranging from 450°C to 700°C, and N2 flow rates ranging 200 to 1,200 cm3 min−1 were investigated. The microwave power, reaction temperature, and N2 flow rates are directly proportional to hydrogen and inversely proportional to biochar yield. The characteristics of the biochar were analysed using CHNOS, scanning electron microscopy, and Brunauer–Emmett–Teller results. Central composite design was used to optimize the number of experiment. The optimized experimental datasets were used as inputs for developing a model that can predict the weight of hydrogen and biochar from microwave pyrolysis of OPF. A multilayer feedforward neural network model was developed by comparing the performance of 11 different types of backpropagation training algorithms. The 11 training algorithms belong to six classes, which are (a) additive momentum, (b) self‐adaptive learning rate, (c) resilient backpropagation, (d) conjugate gradient, (e) quasi‐Newton, and (f) Bayesian regulation (BR). Best performing training algorithm was selected based on the lowest error values computed. The objective of this research is to identify the most suitable training algorithm for this process. Levenberg–Marquardt and BR exhibited very good performance. BR exhibited best performance with lowest error values in predicting weight of hydrogen and biochar. The testing data error values for hydrogen weight is 0.216 root mean square error (RMSE), 0.144 mean absolute error (MAE), and 0.020 mean absolute percentage error (MAPE), and biochar is 0.886 RMSE, 0.489 MAE, and 0.019 MAPE. The optimum number of hidden neuron for this algorithm is 19 with mean square error 0.23. The biochar exhibits porous structure suggesting possible land application.

Artificial neural network modeling of performance, emission, and vibration of a CI engine using alumina nano-catalyst added to diesel-biodiesel blends

In recent years, added nano-catalysts to fuels has improved their thermo-physical properties. In present study, the alumina as additive with concentrations of 30, 60, and 90 ppm were added to B5 and B10 blends for evaluation of the engine performance, emissions, and vibration levels. An ANN model based on standard back-propagation learning algorithm for the engine was developed. Multi-layer perception network (MLP) was used for a non-linear mapping between the input and target parameters. The input or independent parameters were fuel blend, engine speed, fuel density, fuel viscosity, LHV, intake manifold pressure, fuel consumption, exhaust gas temperature, oxygen contained in exhaust gases, oil temperature, relative humidity, and ambient air pressure. Whereas, the target parameters separately were engine power, torque, emissions of CO, CO2, UHC, NO, RMS and Kurtosis of engine’s vibration. The results for optimum ANN model showed, the training algorithm of back-propagation with 25-25 neurons in hidden layers (logsig-logsig) is able to predict different parameters of engine for different conditions. The corresponding R-values for training, validation and testing were 0.9999, 0.9994 and 0.9995, respectively. The performance and accuracy of the proposed ANN model was completely satisfactory.

Predicting the Sorption of Mn2+ ion from Wastewater onto Zinc Chloride Activated Sawdust using Artificial Neural Network (ANN)

Activated carbon from sawdust was prepared and its textural properties was evaluated using Fourier transform infra-red (FTIR) and scanning electron microscope (SEM) to determine the presence of functional groups and visualize its microstructural arrangement in other to ascertain its potential for the removal of Mn 2+ ion from wastewater. Batch experimental technique was then employed to evaluate the effects of adsorption variables, namely; initial metal ion concentration, adsorbent dose, pH, and contact time on the sorption of Mn 2+ ion onto acid activated sawdust. Thereafter, statistical design of experiment (DOE) using central composite design was employed to randomized the levels of selected input parameters in other to produce an experimental design matrix. To model and predict the sorption of Mn 2+ ion onto acid activated sawdust, artificial neural network (ANN) was employed. To apply the neural network, the input data were first normalized to avoid the problems of weight variation. Thereafter, different training algorithm and hidden neurons were selected and tested to ascertain the optimum number of hidden neuron and the best training algorithm that will produce the most accurate network for predicting the sorption of Mn 2+ ion. Results obtained show that Levenberg Marquardt Back Propagation training algorithm with 10 hidden neurons in the input and output layer with tangent sigmoid transfer function produced the most accurate prediction network. In addition, artificial neural network gave a strong agreement between the experimental and predicted sorption efficiency of Mn 2+ ion with coefficient of determination (R 2 ) value of 0.9893. Keywords: Activated sawdust, Artificial neural network, central composite design, Levenberg Marquardt Back Propagation training algorithm DOI : 10.7176/JEES/9-10-15 Publication date :October 31 st 2019

New Artificial Neural Network Models for Bio Medical Image Compression

This article presents an image compression method using feed-forward back-propagation neural networks (NNs). Marked progress has been made in the area of image compression in the last decade. Image compression removing redundant information in image data is a solution for storage and data transmission problems for huge amounts of data. NNs offer the potential for providing a novel solution to the problem of image compression by its ability to generate an internal data representation. A comparison among various feed-forward back-propagation training algorithms was presented with different compression ratios and different block sizes. The learning methods, the Levenberg Marquardt (LM) algorithm and the Gradient Descent (GD) have been used to perform the training of the network architecture and finally, the performance is evaluated in terms of MSE and PSNR using medical images. The decompressed results obtained using these two algorithms are computed in terms of PSNR and MSE along with performance plots and regression plots from which it can be observed that the LM algorithm gives more accurate results than the GD algorithm.

Electric Forecasting using Nature Inspired Optimization Techniques

Now a day, there exists huge competition among power industries in terms of fulfilling various customers’ electrical needs. Reliable and quality power supply is no doubt a basic need for all power consumers. Moreover, planning & operation engineers also targets for proper unit commitment, economic power dispatch, etc., and highly depend upon good power system planning. Therefore, Electric Forecasting (EF) is a major criterion for power engineers. In this manuscript, Artificial Neural Network (ANN), being a well established tool for modeling non-linear and black box systems, is used to forecast hydro generation power plant, energy met and peak demand of India. Furthermore, in this competitive world, ANN model is further optimized using genetic algorithm (GA) and particle swarm optimization (PSO) to explore accurate forecasting model with minimal amount of error. These optimization methods explore highly diversified search area, resulting in more accurate forecasting results in comparison to ANN when trained with standard back propagation training algorithm.

A novel approach for efficient crop yield prediction

Crop yield prediction is one of the challenging task in agricultural domain. Extensive research in agricultural domain has been carried out to predict better crop yield using the machine learning algorithm Artificial Neural Network (ANN) and statistical model Multiple Linear Regression (MLR). This article examines the intrinsic relationship between MLR and ANN. A hybrid MLR-ANN model has been proposed in this research work for efficient crop yield prediction. The proposed hybrid model is modeled to analyze the prediction accuracy when MLR intercept and coefficients were applied to initialize the ANN’s input layer weights and bias. Feed Forward Artificial Neural Network with Back Propagation training algorithm was used for predicting accurate paddy crop yield. In conventional ANN model, the weights and bias of input and hidden layer are initilized randomly. This hybrid MLR-ANN model, instead of random weights and bias initialization, the input layer weights and bias are initialized by using MLR’s coefficients and bias. The hybrid model prediction accuracy is compared with ANN, MLR, Support Vector Regression (SVR), k-Nearest Neighbour (KNN) and Random Forest (RF) models by using performance metrics. The computational time for both hybrid MLR-ANN and conventional ANN was calculated. The results show that the proposed hybrid MLR-ANN model gives better accuracy than the conventional models.

Performance evaluation of gasification system efficiency using artificial neural network

Gasification is one of the thermo-chemical energy conversion processes with energy self-sufficiency, recoverability and controllable efficiency when compared to combustion and pyrolysis. Gasification process is carried out using different systems (gasifiers), different types of fuels and process conditions, and these factors determine the efficiency of the overall process. The performance of a gasifier is evaluated by some parameters such as Cold Gas Efficiency (CGE), Carbon Conversion Efficiency (CCE), gas yield, gas composition, and lower heating value (LHV) of gas. To understand how efficient a gasifier is, several experiments are needed, but conducting these experiments is time consuming and capital intensive, and the information is vital in energy production plants. Meanwhile, a model that could accurately predict the aforementioned parameters irrespective of the type of gasifier, fuel, and operating conditions is imperative for some time conditions. In this study, Levenberg-Marquardt (LM) back-propagation and Bayesian Regularisation (BR) training algorithms for an Artificial Neural Networks (ANN) were used to study a dataset containing 315 experimental data of biomass, coal, and blends of biomass and coal from various gasifiers and process conditions. Eleven input variables were used in the study, and the result shows that the Mean Square Error (MSE) of the BR algorithm was higher than that of the L-M algorithm. To reduce the MSE, techniques called Input Variables Representation Technique with Visual Inspection method (IVRT-VIM) and Output Variables Representation Technique with Visual Inspection Method (OVRT-VIM) were developed, and their applications produced smaller MSE and an R2 of between 79 – 98% and 95–96%, respectively. Further, the results of the sensitivity analysis revealed that carbon (% amount) is the most important input parameter affecting the outputs, and with sum of the Squares of the Partial Derivatives (SSD) value of 1.18 for the CGE prediction.

Automatic Recovery Estimation of Degraded Soils by Artificial Neural Networks in Function of Chemical and Physical Attributes in Brazilian Savannah Soil

ABSTRACTThe Oxisols is predominant in 54% of Brazilian territories and characterized by high weathering, relatively low chemical properties, and adequate structure. This study aimed to analyze the Oxisols through an Artificial Neural Network (ANN) with the purpose of estimating its recovery in function to soil chemical and physical attributes. The chemical attributes considered were: pH, cation exchange capacity (CEC), base saturation (V%), phosphorus (P), magnesium (Mg2+), and potassium (K+) and for the physical attributes, bulk density, soil porosity and soil resistance to penetration. The ANN used in this study is the Multilayer Perceptron (MLP), composed of three layers, input, intermediate and the output and with backpropagation training algorithm (supervised training). The intermediate layer is composed by 10 neurons and the layer of exit by 1 neuron, which has a function of informing the levels of chemical recovery (high, medium and low chemical attributes of the soil) and soil physics (recovered, partially recovered or not recovered). From the results obtained by ANN showed that the network reached an adequate training, with low mean square error (MSE). Therefore, ANN is a powerful and automatic alternative for the recovery estimation of degraded soils.

Accurate and cost-effective prediction of HBsAg titer in industrial scale fermentation process of recombinant Pichia pastoris by using neural network based soft sensor.

In the current work, the attempt was made to apply best-fitted artificial neural network (ANN) architecture and the respective training process for predicting final titer of hepatitis B surface antigen (HBsAg), produced intracellularly by recombinant Pichia pastoris Mut+ in the commercial scale. For this purpose, in large-scale fed-batch fermentation, using methanol for HBsAg induction and cell growth, three parameters of average specific growth rate, biomass yield, and dry biomass concentration-in the definite integral form with respect to fermentation time-were selected as input vectors; the final concentration of HBsAg was selected for the ANN output. Used dataset consists of 38 runs from previous batches; feed-forward ANN 3:5:1 with training algorithm of backpropagation based on a Bayesian regularization was trained and tested with a high degree of accuracy. Implementing the verified ANN for predicting the HBsAg titer of the five new fermentation runs, excluded from the dataset, in the full-scale production, the coefficient of regression and root-mean-square error were found to be 0.969299 and 2.716774, respectively. These results suggest that this verified soft sensor could be an excellent alternative for the current relatively expensive and time-intensive analytical techniques such as enzyme-linked immunosorbent assay in the biopharmaceutical industry.

Time Series Forecasting Using Neural Networks: Are Recurrent Connections Necessary?

Artificial neural networks (NNs) are widely used in modeling and forecasting time series. Since most practical time series are non-stationary, NN forecasters are often implemented using recurrent/delayed connections to handle the temporal component of the time varying sequence. These recurrent/delayed connections increase the number of weights required to be optimized during training of the NN. Particle swarm optimization (PSO) is now an established method for training NNs, and was shown in several studies to outperform the classical backpropagation training algorithm. The original PSO was, however, designed for static environments. In dealing with non-stationary data, modified versions of PSOs for optimization in dynamic environments are used. These dynamic PSOs have been successfully used to train NNs on classification problems under non-stationary environments. This paper formulates training of a NN forecaster as dynamic optimization problem to investigate if recurrent/delayed connections are necessary in a NN time series forecaster when a dynamic PSO is used as the training algorithm. Experiments were carried out on eight forecasting problems. For each problem, a feedforward NN (FNN) is trained with a dynamic PSO algorithm and the performance is compared to that obtained from four different types of recurrent NNs (RNN) each trained using gradient descent, a standard PSO for static environments and the dynamic PSO algorithm. The RNNs employed were an Elman NN, a Jordan NN, a multirecurrent NN and a time delay NN. The performance of these forecasting models were evaluated under three different dynamic environmental scenarios. The results show that the FNNs trained with the dynamic PSO significantly outperformed all the RNNs trained using any of the other algorithms considered. These findings highlight that recurrent/delayed connections are not necessary in NNs used for time series forecasting (for the time series considered in this study) as long as a dynamic PSO algorithm is used as the training method.

Performance Analysis of ANN Based three-phase four-wire Shunt Active Power Filter for Harmonic Mitigation under Distorted Supply Voltage Conditions

This paper presents the design and implementation of three-phase four-wire shunt active power filter (SAPF). It consists of insulated gate bipolar transistors, IGBT based current-controlled voltage source inverter (CC-VSI), series coupling inductor and self-supported DC bus. Power electronics based converters and non-linear loads generate waveform-driven power quality issue as harmonics. Three-phase four-wire SAPF mitigates harmonics, compensates for reactive power, neutral current and power factor correction. Conventionally, the positive sequence detection control strategy using phase-locked loop (PLL) is applied as the synchronizing unit vector element to generate reference source currents. Conventional controller tuning process is difficult and fails to perform satisfactorily under supply voltage variation conditions. In this paper, Levenberg-Marquardt back propagation training algorithm based artificial neural network (ANN) controller is proposed to regulate DC link voltage due to its self-adapting and rapid calculation characteristics that allow the controller to handle high nonlinearity and uncertainty in a non-linear system. Weights of a neuron are adapted to minimize total harmonic distortion (THD) of source current under the step, ramp, time series amplitude variation and frequency and amplitude of modulation conditions. The proposed system is modelled in MATLAB/SIMULINK environment and laboratory prototype with dSpace1104 control card is developed. Experimentation results validate the simulated results of the proposed scheme under supply voltage variations for three-phase four-wire distribution system.

Calcareous Soil as a Promising Adsorbent to Remove Fluoride from Aqueous Solution: Equilibrium, Kinetic and Thermodynamic Study

In this work, the feasibility of employing calcareous soil to remove fluoride ions from its aqueous solutions was investigated under batch mode. The influence of solution pH, sorbent dose, initial fluoride concentration, contact time, stirring rate and temperature on the removal process were investigated. The equilibrium adsorption data were analyzed using Langmuir, Freundlich and Temkin isotherm models. The kinetics of fluoride ion was discussed by pseudo-first-order, pseudo-second-order and intraparticle diffusion models. It was shown that the adsorption of fluoride ions could be described by the pseudo-second-order kinetic model. Thermodynamic parameters such as Gibbs free energy (rG o ), enthalpy (rH o ) and the entropy change of sorption (rS o ) have also been evaluated and it has been found that the adsorption process was spontaneous, feasible and exothermic in nature. A six-layered feed forward neural network with back propagation training algorithm was developed using thirty-one experimental data sets obtained from laboratory batch study. The ANN predicted results were compared with the experimental results of the laboratory test. Regeneration study indicates that about 81.4% fluorides can regenerate from the adsorbent. It was concluded that calcareous soil has potential for application as an effective adsorbent for removal of fluoride ions from aqueous solution

Simulation and prediction of optical characterization of casting (Adamantan-Fulgide) thin films using (CTANNs) approach

For reinforcement, the photochromic field and the cooperation between the theoretical and experimental branches of physics, the computational, theoretical artificial neural networks (CTANNs) and the resilient back propagation (R[Formula: see text]) training algorithm were used to model optical characterizations of casting (Admantan-Fulgide) thin films with different concentrations. The simulated values of ANN are in good agreement with the experimental data. The model was also used to predict values, which were not included in the training. The high precision of the model has been constructed. Moreover, the concentration dependence of both the energy gaps and Urbach’s tail were, also tested. The capability of the technique to simulate the experimental information with best accuracy and the foretelling of some concentrations which is not involved in the experimental data recommends it to dominate the modeling technique in casting (Admantan-Fulgide) thin films.

Analysis and diagnosis of PEM fuel cell failure modes (flooding & drying) across the physical parameters of electrochemical impedance model: Using neural networks method

The objective of this work is to define and implement a method of artificial neural network to create an optimal impedance model of the proton exchange membrane fuel cell (PEMFC) which considers the electrochemistry and the mass transfer theory, which are used to analyze and diagnose PEM fuel cell failure modes (flooding & drying) across the physical parameters of the electrochemical impedance model. For this, we have based on the neural network technique for the calculation and estimation of various constituents parameters of this model. The Multi-Layer-Perceptron through back-propagation training algorithms shows satisfactory performance with the regard of parameter prediction. Furthermore, the neural network method applied to the impedance model is valid and valuable, which used to estimate the physical parameters of the electrochemical impedance model of the fuel cell (PEMFC) in both cases; flooding and drying of the fuel cell heart. The novelty of our work is summed up in the demonstration of the existence in a simple and uncomplicated way that allows the knowledge of the state of health of the PEMFC.

Prediction of Hourly Cooling Energy Consumption of Educational Buildings Using Artificial Neural Network

Predicating the required building energy when it is in the design stage and before being constructed considers a crucial step for in charge people. Hence, the main aim of this research is to accurately forecast the needed building cooling energy per hour for educational buildings at University of Technology in Iraq. For this purpose, the feed forward artificial neural network (ANN) has been selected as an efficient technique to develop such a predication system. Firstly, the main building parameters have been investigated and then only the most important ones were chosen to be used as inputs to the ANN model. However, due to the long time period that is required to collect actual consumed building energy in order to be employed for ANN model training, the hourly analysis program (HAP), which is a building simulation software, has been utilized to produce a database covering the summer months in Iraq. Different training algorithms and range of learning rate values have been investigated, and the Bayesian regularization backpropagation training algorithm and 0.05 learning rate were found very suitable for precise cooling energy prediction. To evaluate the performance of the optimized ANN model, mean square error (MSE) and correlation coefficient (R) have been adopted. The MSE and R indices for the predication results proved that the optimized ANN model is having a high predication accuracy with 5.99*10-6 and 0.9994, respectively.