Feed Forward Back Propagation Method Research Articles

Precipitation Forecast with Artificial Neural Networks Method

Events in the atmosphere from past to present – wind, precipitation, humidity, temperature – have almost always been the subject of research to create a forecast in regions. The rapid development of the technological field in terms of software and hardware brings methods and techniques to be used in research. One of them is Artificial Neural Networks. In this study, precipitation data were estimated using the Feed Forward Backpropagation method of Artificial Neural Networks method using past data of meteorological parameters, and they were compared with the data of multiple linear regression analysis. Based on these models, six different models were studied, and regression and performance evaluations were made. While the error average of multiple linear regression is 0.2413, this value is 0.076 in artificial neural networks, and the correlation average for both is 0.90. As a result of this study, the best model has a coefficient of determination of 0.95 and an error value of 0.18 in multiple linear regression, as well as a coefficient of certainty of 0.99 and an error value of 0.0438 in artificial neural networks; It has been understood that the 1st model, which has 6 data sets as the input layer, exhibits the best performance.

Performance assessment of the indigenous ceramic UF membrane in bioreactor process for highly polluted tannery wastewater treatment

The present study evaluates the performance of an indigenously developed ceramic ultrafiltration (UF) membrane in a lab-scale membrane bioreactor (MBR) process to treat real tannery effluent with varying organic loading (1500-6000mg/L). UF membrane was prepared by the coating of bentonite clay on tubular clay-alumina macroporous support. The membrane surface was characterized by scanning electron microscopy, Fourier transform infrared spectroscopy, X-ray diffraction, X-ray photoelectron spectroscopy, BET adsorption-desorption, contact angle measurement, and atomic force microscopy. In the side-stream MBR process, sewage sludge from a local sewage treatment plant was used as an activated sludge source with a constant sludge retention time of 30days. Membrane filtration was performed in cross-flow mode using a single-channel membrane module. Artificial neural network (ANN) modeling tool was used to analyze the influence of various independent input variables, namely, the hydraulic retention time (4-10h), mixed liquor suspended solid (MLSS) concentration (2-8g/L), and influent COD concentration (1500-6000mg/L) on COD removal (%) with feed-forward backpropagation method. Membrane study was done at a transmembrane pressure of 4.3bar and feed flow rate of 7.5 L/min to observe the flux declination and fouling of the UF membrane with time. Average COD and BOD concentrations obtained in the treated effluent were 147.56 and 31mg/L, respectively, and chromium concentration was < 0.1mg/L; thus, treated effluent quality was found to be suitable for industrial recycling purposes apart from the safe environmental discharge. An in-depth study was undertaken to understand the removal mechanism in the MBR process, nature and extent of membrane fouling, changes in the morphology of the UF membrane, surface wettability, and surface topology by detailed surface characterization of the membrane pre- and post-filtration.

Evaluation of Physicothermal Properties of Silicone Oil Dispersed with Multiwalled Carbon Nanotubes and Data Prediction Using ANN

The researchers wanted to see whether MWCNTs changed the physicothermal properties of solar thermal working fluids. Assessing thermal properties is vital for solar thermal efficiency. Lubricant contains silicone oil resurfaced. It contains 0.25, 0.5, 0.75, and 1.0% multiwalled carbon nanotubes. Before dispersion in thermic fluids, nanomaterials must be properly surface modified. Between 100°C and 300°C, a fluid’s thermal conductivity and specific heat physical characteristics like viscosity and density may be inferred from data collected between 50°C and 150°C. Thermal conductivity increases by 15% to 20% when carbon nanotubes are dispersed. The pressure drop is minimal at 0.5 percent weight fraction, demonstrating the suitability of nanofluids in closed loop systems. The characteristics are forecasted using feed-forward backpropagation method and GRNN, and the best of them is selected for prediction. In this research, hidden layer neurons and factors are examined.

OPTIMIZATION USING RESPONSE SURFACE METHODOLOGY AND ARTIFICIAL NEURAL NETWORK IN GEARED DRILLING

Drill on different layered composite causes to force the layers ahead of it, creating unacceptable delamination on the leaving side. The current work describes the influence of different process parameters like spindle speed, feed, diameter of drill bit and addition of filler material on delamination damage on carbon and jute hybrid fiber composites during drilling operation. The experimental output for delamination is optimized by RSM-Response Surface Methodology and ANN-Artificial Neural Network Model. The experimental prediction model was established by considering DOE (design of experiments) of three levels performed with drilling operation by varying above said process parameters The optimum values for minimized delamination damage conditions found to be at (J-C-C-J) +5g of filler (fly-ash) composite. This is drilled with 8.5mm diameter drill bit with a feed of 0.08mm/rev at 875rpm speed. From the theoretical results it is recognized that cutting-speed and filler-material have much influencing factors on responses (delamination), and their individual contribution in an order of 47.25% and 47.32% respectively. By using Box-Behnken design RSM model is developed, with a feed-forward back-propagation method to develop the predictive ANN model which consists of 15 neurons in its hidden layer along with ANN Model. Here ANN Results (R2=0.99and RMSE=1.99) showed that the developed model is performing better to predict content of delamination when compare to RSM results (R2=0.97and RMSE=2.24).

Development of traffic noise prediction model for major arterial roads of tier-II city of India (Surat) using artificial neural network

Abstract In the issue of expanding noise levels the world over, road traffic noise is main contributor. The investigation of street traffic noise in urban communities is a significant issue. Ample opportunity has already passed to understand the significance of noise appraisal through prediction models with the goal that assurance against street traffic noise can be actualized. Noise predictions models are utilized in an increasing range of decision-making applications. This study’s main objective is to assess ambient noise levels at major arterial roads of Surat city, compare these with prescribed standards, and develop a noise prediction model for arterial roads using an Artificial Neural Network. The feed-forward back propagation method has been used to train the model. Models have been developed using the data of three roads separately, and one final model has also been developed using the data of all three roads. Among the prediction in three arterial roads, the predicted output result from the model of Adajan-Rander showed a better correlation with a mean squared error (MSE) of 0.789 and R2 value of 0.707. But with the combined model, there is a slight deterioration in mean squared value (MSE) 1.550, with R2 not getting changed much significantly, i.e., 0.755. However, the combined model’s prediction can be adopted due to the variety of data used in its training.

Prediction of Mechanical Properties of Fe 415 Steel in Hot Rolling Process Using Artificial Neural Network

The purpose of the work is to predict the strength of the rolled product using artificial neural network (ANN). A network model is developed to predict the strength of rolled Fe 415 steel. Manual test result data are collected from testing laboratory for output/response parameters, viz. UTS, YS and % elongation by taking FST, WP, WFR and LHT as input parameters. Then, ANN tool is used to train the network using feedforward backpropagation method and validate the network for best fit. The network model is tested with 70% of data values, and probability of prediction is checked. Predicted value of parameters is compared with experimental values via percentage deviation and confirmative analysis. In the result, the predicted value of strength is found to be comparable and satisfactory with experimental values with deviation value as − 0.2250. The suggested ANN model can also be utilized for the prediction of properties of other processes.

Successful Implementation of Artificial Intelligence and Machine Learning in Multiphase Flow Smart Proxy Modeling: Two Case Studies of Gas-Liquid and Gas-Solid CFD Models

It is almost impossible to solve the modern fluid flow problems without the use of Computational Fluid Dynamics (CFD). In petroleum industry, flow simulations assist engineers to develop the most efficient well design and it is essential to understand the multiphase flow details. However, despite the high accuracy, performing the numerical simulation fall short in providing the required results in timely manner. This article presents two case studies of Smart Proxy Models (SPM) utilizing artificial intelligence (AI) and Machine Learning (ML) techniques to appraise the behavior of the chaotic system and predict the dynamic features including pressure, velocity and the evolution of phase fraction within the process at each time-step at a much lower run time. Proposed cases concentrate on 2-D dam-break and 3-D fluidized bed problems, using OpenFOAM and MFiX, CFD software applications, respectively. This paper focuses on building and improving the artificial neural network (ANN) models characterized by feedforward back propagation method and Levenberg-Marquardt algorithm (LMA). Each case study contains multiple scenarios to gradually enhance the model capabilities to forecast the dynamic parameters. Results for both cases indicate that 8-10 hours of computational time for running CFD simulation, reduces to a few minutes when is done by developed AI-based models along with less than 10% error for entire process.

An Improved Vehicle Parking Mechanism to reduce Parking Space Searching Time using Firefly Algorithm and Feed Forward Back Propagation Method

An issue of finding parking space is a serious concern in some locations, especially shopping complexes, hospitals, buildings, malls and other premises requires large parking space for vehicles. The parking scenario becomes worst on special occasions like festivals, fiesta and weekends. The conventional developed techniques using installation of sensors in parking zone for parking vehicles becomes costly. Therefore, a reliable method is required that works for long time to manage traffic congestion that occurs while searching for parking space. In this paper, a novel method is proposed for parking vehicles using metaheuristic approaches. The developed approach provides consistent results considering two parameters, parking efficiency and parking space search time. The parking efficiency is improved and parking space search time is reduced using the Firefly Algorithm (FA) and Feed Forward Back Propagation Neural Network (NN) approach.

Breast Cancer Classification Using Artificial Neural Network and Feature Selection

Breast cancer is still one of the leading causes of death in the world. Prevention can be done if the cancer can be recognized early on whether the cancer is malignant or benign. In this study, a comparison of malignant and benign cancer classifications was performed using two artificial neural network methods, which are the Feed-Forward Backpropagation method and the Elman Recurrent Neural Network method, before and after the feature selection of the data. The result of the study produced that Feed-Forward Backpropagation method using 2 hidden layers is better after the feature selection was performed on the data with an accuracy value of 99,26%.

Artificial neural network modeling and genetic algorithm optimization of process parameters in fluidized bed drying of green tea leaves

AbstractThe present study involved integration of artificial neural network (ANN) with genetic algorithm (GA) for predicting the optimized process parameters required for fluidized bed drying of green tea leaves. It had a layer each for input and output with linear activation function and two hidden layers with a sigmoid function. The feed forward back propagation method was used to train the developed model. The input parameters used by ANN for generalizing the drying process were temperature (50–80°C) and air flow velocity (7–9.5 m/s), and the output parameters were drying time, total color difference (TCD), and total phenolic content (TPC). The weights and bias values of trained ANN were used by GA to estimate the fitness function, which maximizes the TPC and minimizes the drying time and TCD. The optimum process condition of independent variables (80°C and 9 m/s) obtained from the hybrid ANN/GA was validated, and agreeable relationship between actual and predicted values with relative standard deviation (SD) of 5.7, 0.46, and 0.22 was found for drying time, TCD and TPC of dried leaves, respectively. Hence, under this optimal drying condition, best quality green tea can be obtained within the limits defined.Practical applicationsDrying is a popular unit operation being preferred to prolong shelf life of cash crops. Air drying is predominantly used at any given scale due to its economic feasibility. However, the quality parameters get compromised during the process of drying. Hence, an optimal drying condition is necessary to have dried product with intact nutritional and functional activity, alongside consumer acceptability. This was attempted using ANN and GA. Artificial neural network and genetic algorithm can estimate the optimum fluidized bed drying condition to have intact physical and chemical properties of green tea leaves. The results indicated that the developed ANN/GA drying model can efficiently estimate the values of quality parameters of dried green tea leaves, and also identify the optimal drying conditions for any new data.

DIAGNOSIS OF CARDIAC ABNORMALITY USING HEART SOUND

Heart sound (HS) analysis or auscultation is a standout amongst the most simple, non-invasive and costless methods used to evaluate heart health and is one of the basic and foremost routine of a doctor while reviewing a patient. Detecting cardiac abnormality by auscultation demands a physician’s experience and even then there is a high scope of committing error. In this paper, a low cost electronic stethoscope is built to acquire HS in a novel manner by taking one from each ventricular and auricular area and superimposed, to get a resultant signal of both distinct lub-dub sound. Then, a light, fast and low computation speed beat track method followed by wavelet reconstruction is presented for correct detection of S1 and S2. It is done without ECG reference, and can be used satisfactorily on both normal and pathological HSs. Moreover, heartbeats can be identified in both de-noised and noised environment as it is independent of external disturbances. Significant features are extracted from the resultant HSs with detected S1 and S2 and feed-forward back propagation method. It is used to classify the HS nature into normal and pathological. This algorithm has been implemented on 24 pairs of HSs, extracted from 24 patients of 15 pathological and nine normal subjects and the classification yields a result of 91.7% accuracy with 81.8% sensitivity. The overall performance suggests a good performance to cost ratio. This system can be used as first diagnosis tool by the medical professionals.

Optimization and Prediction of Motorcycle Injection System Performance with Feed-Forward Back-Propagation Method Artificial Neural Network (ANN)

This research studied the use of Artificial Neural Network (ANN) using feed-forward back-propagation model to optimize and predict the performance of a motorcycle fuel injection systems of gasoline. The parameters such as speed, throttle position, ignition timing and injection timing is used as the input parameters. While the parameters of fuel consumption and engine torque is used as the output layer. Lavenberg-Marquardt model type with train function tanh sigmoid and 25 neurons number is used to generate the target value and the desired output. Variation of ignition timing as optimization variable in a wide range of speed and throttle position is used in experimental tests. ANN is used to investigate the prediction of performance motorcycle engines and compared with the test results. Results showed that the operation of ANN in predicting engine performance is very good. From the test results obtained a smooth contour MAP compared to the initial state. The prediction result and performance test show a good correlation in small error value of training and test that is regression with range 0.98-0.99, mean relative error with range 0.1315-0.4281% and the root mean square error with range 0.2422-0.9754%. This study shows that the feed-forward back propagation on ANN model can be used to predict accurately the performance of a motorcycle engine injection system.

ASSESSMENT OF THE ARTIFICIAL NEURAL NETWORKS TO ‎GEOMORPHIC MODELLING OF SEDIMENT YIELD FOR ‎UNGAUGED CATCHMENTS, ALGERIA

Knowledge of sediment yield and the factors controlling it provides useful ‎information for estimating ‎erosion intensities within river basins. The objective of ‎this study was to build a model from which ‎suspended sediment yield could be ‎estimated from ungauged rivers using computed sediment yield and ‎physical ‎factors. Researchers working on suspended sediment transported by wadis in the ‎Maghreb are ‎usually facing the lack of available data for such river types. Further ‎study of the prediction of sediment ‎transport in these regions and its variability is ‎clearly required. In this work, ANNs were built between ‎sediment yield ‎established from longterm measurement series at gauging stations in Algerian ‎catchments and ‎corresponding basic physiographic parameters such as rainfall, ‎runoff, lithology index, coefficient of ‎torrentiality, and basin area. The proposed ‎Levenberg-Marquardt and Multilayer Perceptron algorithms to ‎train the neural ‎networks of the current research study was based on the feed-forward ‎backpropagation ‎method with combinations of number of neurons in each hidden ‎layer, transfer function, error goal. ‎Additionally, three statistical measurements, ‎namely the ‎root mean square error (RMSE), ‎the coefficient of ‎determination (R²), ‎and the efficiency factor (EF)‎ have been reported for ‎examining the forecasting ‎‎accuracy of the developed model.‎ Single plot displays of network outputs with ‎respect to targets for training ‎have provided good performance results and good ‎fitting . Thus, ANNs were a promising method for ‎predicting suspended sediment ‎yield in ungauged Algerian catchments.‎

ASSESSMENT OF THE ARTIFICIAL NEURAL NETWORKS TO GEOMORPHIC MODELLING OF SEDIMENT YIELD FOR UNGAUGED CATCHMENTS, ALGERIA

Knowledge of sediment yield and the factors controlling it provides useful ‎information for estimating ‎erosion intensities within river basins. The objective of ‎this study was to build a model from which ‎suspended sediment yield could be ‎estimated from ungauged rivers using computed sediment yield and ‎physical ‎factors. Researchers working on suspended sediment transported by wadis in the ‎Maghreb are ‎usually facing the lack of available data for such river types. Further ‎study of the prediction of sediment ‎transport in these regions and its variability is ‎clearly required. In this work, ANNs were built between ‎sediment yield ‎established from longterm measurement series at gauging stations in Algerian ‎catchments and ‎corresponding basic physiographic parameters such as rainfall, ‎runoff, lithology index, coefficient of ‎torrentiality, and basin area. The proposed ‎Levenberg-Marquardt and Multilayer Perceptron algorithms to ‎train the neural ‎networks of the current research study was based on the feed-forward ‎backpropagation ‎method with combinations of number of neurons in each hidden ‎layer, transfer function, error goal. ‎Additionally, three statistical measurements, ‎namely the ‎root mean square error (RMSE), ‎the coefficient of ‎determination (R²), ‎and the efficiency factor (EF)‎ have been reported for ‎examining the forecasting ‎‎accuracy of the developed model.‎ Single plot displays of network outputs with ‎respect to targets for training ‎have provided good performance results and good ‎fitting . Thus, ANNs were a promising method for ‎predicting suspended sediment ‎yield in ungauged Algerian catchments.‎

ASSESSMENT OF THE ARTIFICIAL NEURAL NETWORKS TO ‎GEOMORPHIC MODELLING OF SEDIMENT YIELD FOR ‎UNGAUGED CATCHMENTS, ALGERIA

Knowledge of sediment yield and the factors controlling it provides useful ‎information for estimating ‎erosion intensities within river basins. The objective of ‎this study was to build a model from which ‎suspended sediment yield could be ‎estimated from ungauged rivers using computed sediment yield and ‎physical ‎factors. Researchers working on suspended sediment transported by wadis in the ‎Maghreb are ‎usually facing the lack of available data for such river types. Further ‎study of the prediction of sediment ‎transport in these regions and its variability is ‎clearly required. In this work, ANNs were built between ‎sediment yield ‎established from longterm measurement series at gauging stations in Algerian ‎catchments and ‎corresponding basic physiographic parameters such as rainfall, ‎runoff, lithology index, coefficient of ‎torrentiality, and basin area. The proposed ‎Levenberg-Marquardt and Multilayer Perceptron algorithms to ‎train the neural ‎networks of the current research study was based on the feed-forward ‎backpropagation ‎method with combinations of number of neurons in each hidden ‎layer, transfer function, error goal. ‎Additionally, three statistical measurements, ‎namely the ‎root mean square error (RMSE), ‎the coefficient of ‎determination (R²), ‎and the efficiency factor (EF)‎ have been reported for ‎examining the forecasting ‎‎accuracy of the developed model.‎ Single plot displays of network outputs with ‎respect to targets for training ‎have provided good performance results and good ‎fitting . Thus, ANNs were a promising method for ‎predicting suspended sediment ‎yield in ungauged Algerian catchments.‎

Comparison of Artificial Neural Network Architecture in Solving Ordinary Differential Equations

This paper investigates the solution of Ordinary Differential Equations (ODEs) with initial conditions using Regression Based Algorithm (RBA) and compares the results with arbitrary- and regression-based initial weights for different numbers of nodes in hidden layer. Here, we have used feed forward neural network and error back propagation method for minimizing the error function and for the modification of the parameters (weights and biases). Initial weights are taken as combination of random as well as by the proposed regression based model. We present the method for solving a variety of problems and the results are compared. Here, the number of nodes in hidden layer has been fixed according to the degree of polynomial in the regression fitting. For this, the input and output data are fitted first with various degree polynomials using regression analysis and the coefficients involved are taken as initial weights to start with the neural training. Fixing of the hidden nodes depends upon the degree of the polynomial. For the example problems, the analytical results have been compared with neural results with arbitrary and regression based weights with four, five, and six nodes in hidden layer and are found to be in good agreement.

Prediction of residual stresses in gas arc welding by back propagation neural network

This paper presents the development of a back propagation neural network model for the prediction of maximum residual stresses produced in gas metal arc welding process. The model is based on results obtained from finite element models. The thickness of the plate, electrode size, welding speed, current/voltage intensity have been considered as the input parameters and the maximum residual stresses due to welding as output parameters in the development of the model. The Levenberg–Marquardt method as a feed forward back propagation method was used in this investigation. The neural network predictions were then compared with the finite element results for accuracy, and the comparison showed that the results obtained from neural network model were sufficiently accurate in predicting the residual stresses.

Observations on the application of artificial neural network to predicting ground motion measures

Application of the artificial neural network (ANN) to predict pseudospectral acceleration or peak ground acceleration is explored in the study. The training of ANN model is carried out using feed-forward back-propagation method and about 600 records from 39 California earthquakes. The statistics of the residuals or modeling error for the trained ANN-based models are almost the same as those for the parametric ground motion prediction equations, derived through regression analysis; the residual or modeling error can be modeled as a normal variate. The similarity and differences between the predictions by these two approaches are shown. The trained ANN-based models, however, are not robust because the models with almost identical mean square errors do not always lead to the same predictions. This undesirable behaviour for predicting the ground motion measures has not been shown or discussed in the literature; the presented results, at least, serve to raise questions and caution on this problem. A practical approach to ameliorate this problem, perhaps, is to consider several trained ANN models, and to take the average of the predicted values from the trained ANN models as the predicted ground motion measure.

Flow estimations for the Sohu Stream using artificial neural networks

In this study, daily rainfall–runoff relationships for Sohu Stream were modelled using an artificial neural network (ANN) method by including the feed-forward back-propagation method. The ANN part was divided into two stages. During the first stage, current flows were estimated by using previously measured flow data. The best network architecture was found to utilise two neurons in the input layer (the delayed flows from the first and second days), two hidden layers, and one output layer (the current flow). The coefficient of determination (R2) in this architecture was 81.4%. During the second stage, the current flows were estimated by using a combination of previously measured values for precipitation, temperature, and flows. The best architecture consisted of an input layer of 2 days of delayed precipitation, 3 days of delayed flows, and temperature of the current. The R2 in this architecture was calculated to be 85.5%. The results of the second stage best reflected the real-world situation because they accounted for more input variables. In all models, the variables with the highest R2 ranked as the previous flow (81.4%), previous precipitation (21.7%), and temperature.

Artificial neural network based modelling of desiccant wheel

In this article, the concept of artificial neural network and goal oriented design have been used to propose a computer design tool that can help designers to evaluate performance of desiccant cooling system and behaviour of the desiccant wheel. Based on the experimental observations on desiccant wheel, a neural network model has been developed using a neural network toolbox of MATLAB ® with feed forward back propagation method. The model has been validated against experimental data sets. A number of training algorithms with feed forward back propagation method have been used for the modelling of desiccant wheel to identify a training algorithm with least mean square error (MSE). The performance of all training algorithms has been analyzed and training algorithm trainlm (Levenberg–Marquardt back propagation) is found most suitable for the prediction of outputs which have least mean square error of 0.064462 and 0.007575 for specific humidity and temperatures respectively. The proposed model can predict the specific humidity and temperature at the outlet of desiccant wheel within the range of experimental values.