Cascade Forward Back Propagation Research Articles

A Scale and Rotation Invariant Urdu Nastalique Ligature Recognition Using Cascade Forward Backpropagation Neural Network

In the emerging age of technologies, machines are becoming more and more skilled and capable just like humans. Despite the fact that machines do not have their own intelligence, but still due to advancement in Artificial Intelligence (AI), machines are rapidly advancing. The area of Pattern Recognition (PR) deals with bringing enhancements to identify obscure patterns corresponding to specific classes. Optical Character Recognition (OCR) is a subfield of PR which deals with the recognition of characters. A great work has been done for Japanese, Hindi, Arabic and Chinese scripts, but only a diminutive work has been done for Urdu script. The Urdu language is highly cursive and is written in different calligraphic styles like Naskh, Nastalique, Kofi, Devani and Riqa. The Nastalique font is very calligraphic with aesthetic beauty. The ligature segmentation of Urdu Nastalique is also more difficult as compared to other languages. Urdu Nastalique has some characteristics like stacking of ligatures and cursiveness which makes its ligature segmentation a difficult task. Cursiveness means ligatures are joined together to form a new shape. It contains connected ligatures which makes it more complicated as compared to other languages. The ligature recognition of Urdu text by an OCR is a strenuous task due to variants of scaling, rotation, orientation and font style. In this study, a scale and rotation invariant classifier for Urdu Nastalique OCR is proposed. A combination of scale and location invariant moments is used for feature extraction and the classification is performed using Cascade Forward Backpropagation Neural Network. The model is validated through independent dataset testing and 5-fold cross-validation which gave 96.474% and 96.922% accuracy. The results depict the adaptability of the proposed model due to its high accuracy for recognition of Urdu Nastalique Ligature.

Nonlinear Modeling of the Flux Linkage in 2-D Plane for the Planar Switched Reluctance Motor

This paper proposes a nonlinear flux linkage model in 2-D plane for the planar switched reluctance motor (PSRM). The inputs of the proposed model are the 2-D positions and the current, and the output is the flux linkage. The proposed model is established via a cascade-forward backpropagation neural network (CFNN). The designed CFNN consists of four layers: one input layer, two hidden layers, and one output layer. The first hidden layer has 20 neurons with a tan-sigmoid transfer function, and the second hidden layer has 20 neurons with a log-sigmoid transfer function. The output layer is a pure linear layer. The sample set with 179 755 samples is obtained experimentally in a dSPACE-based PSRM system by applying the dc excitation method. The sample set is divided into three sets. 35% and 30% of the samples are randomly chosen as the training sample set and validation sample set, respectively, and the remaining samples are utilized as the test sample set to assess the generalization performance of the CFNN-based model. According to the results of the test sample set, the maximum relative error is 11.05% and the mean relative error is 0.42% when the current ranges from 1 to 9 A. The CFNN has the capability to build a multi-input nonlinear model. The CFNN-based model is capable of reflecting the variations of flux linkage in 2-D plane caused by manufacturing tolerances. The effectiveness of the CFNN-based model is finally verified.

Crack Width Estimation Using Feed and Cascade Forward Back Propagation Artificial Neural Networks

Automatic crack inspection techniques that limit the necessity of human have the potential to lower the cost and time of the process. In this study, a maximum crack width estimation approach is presented. Seventy nine segments of cracks are used for training the neural networks and twenty six segments are used for examination. The maximum width for each segment is measured using laser scanning microscope and segment image is captured and magnified using the microscope camera in order to obtain the extracted crack profile number of pixels. Feed and cascade forward back propagation artificial neural networks are designed and constructed. The input and output for the networks are the crack width in terms of number of pixels and the maximum estimated crack width respectively. It is shown that, the artificial neural networks technique can effectively be used to estimate the crack width. The feedforward back propagation structure which is designed with two layers and training function TRAINLM gives the best results in examination.

Hot water temperature prediction using a dynamic neural network for absorption chiller application in Indonesia

Weather condition particularly for solar radiation and dry bulb temperature has important role in absorption chiller performance. In this paper hot water temperature prediction in generator inlet of absorption chiller has been conducted under various weather conditions. Dry bulb temperature and global horizontal radiation are selected as predictors. Three artificial neural network (ANN) types including feed forward back-propagation, cascade forward back-propagation, and Elman back propagation models have been investigated for prediction. Moreover, numbers of neuron and time delay effects were analyzed to achieve an accurate prediction. The results show that hot water temperature in generator inlet can be predicted precisely using a feed forward back propagation neural network with the configuration of a three hour delayed input on radiation, current dry bulb temperature, seven neurons, tan-sigmoid transfer function and Bayesian regularization algorithm. The prediction results perform a good agreement between predicted and experimental values. The error resulting from training and validation is 3.1 °C and 2.6 °C with a coefficient of variation at 4.4% and 3.5% respectively.

Demand energy forecasting using genetic algorithm to guarantee safety on electrical transportation system

Demand estimation models are used for energy planning activities. Their primary function is focused on securing energy supply to final users using available resources in generation, transport and interconnection. Long-term planning models typically use non-linear optimization techniques considering an error not exceeding 5%. The reference model used by UPME in Colombia is limited to an average error of 1.6% considering non-linear modeling estimation techniques. However, they are limited in their ability to anticipate uncharacteristic variations in curves or externalities, which increases the probability of an erroneous prediction. Therefore, this research proposes a model to forecast electricity demand using neural networks in order to anticipate non-characteristic variations. The study first documents current methodologies for the prediction of maximum power demand, as well as the current deficiencies in the used forecasts, A new model is then formulated with the application of neural networks using the algorithm Cascade-Forward Back propagation using MATLAB R2017a. During the model comparison process, it was identified that the data obtained reflects the characteristics of demand behavior with an acceptable margin error equal to 0.5%.

Application of response surface methodology and artificial neural network modeling to assess non-thermal plasma efficiency in simultaneous removal of BTEX from waste gases: Effect of operating parameters and prediction performance

Abstract This study aimed to assess the prediction efficiencies of response surface methodology (RSM) and artificial neural network (ANN)-based models in terms of benzene, toluene, ethylbenzene, and xylenes (BTEX) removal from a polluted airstream using non-thermal plasma (NTP). The effect that key elements of the NTP process, including temperature, BTEX concentration, voltage and flow rate, had on the BTEX elimination efficiency was investigated using a central composite RSM design along with three ANN models including Feed-Forward Back Propagation Neural Network (FFBPNN), Cascade-Forward Back Propagation Neural Network (CFBPNN) and Elman-Forward Back Propagation Neural Network (EFBPNN) with the topology of 4-h-1. The RSM and ANN models were statistically compared using some indicators including Sum of Squared Errors (SSE), adjusted R2, determination coefficient (R2), Root Mean Squared Error (RMSE), Absolute Average Deviation (AAD). According to the RSM output, voltage was the most efficient variable with a coefficient proportion of 8.28. Besides, FFBPNN was the best model among the considered ANN models. Also, the R2 achieved for ANN (FFBPNN) and RSM models were 0.9736 and 0.9656 correspondingly. Therefore, it was concluded that the ANN (FFBPNN) represents a powerful tool for modeling the BTEX removal.

Artificial neural network modeling of biosorption process using agricultural wastes in a rotating packed bed

Given the complexity of computational fluid dynamics models and the inaccuracy of semi-empirical models in modeling biosorption process in a rotating packed bed (RPB), an artificial neural network (ANN) based approach was proposed for modeling of biosorption process in the RPB with different biosorbents from agriculture wastes. The experimental data collected from previous studies were used for ANN modeling, and 82% of the data were used for training and 18% of the data were used for testing. The liquid Reynolds number (ReL), average high gravity factor (β), ratio of contact time to maximum contact time (t/tmax), ratio of particle size to bed depth (D/H) and ratio of initial concentration to packing density (C0/ρ) were set as input parameters; while the ratio of the biosorption amount at time t to the maximum biosorption amount (qt/qmax) was used as output parameters for each model. The optimum number of neurons in the hidden layer was determined based on the mean squared errors (MSE) and correlation coefficients (R2) by an optimization procedure. Compared with cascade-forward back-propagation networks (CFBN) and elman back-propagation networks (EBN), feed-forward backpropagation networks (FFBN) gave a lower MSE value and a higher R2 value, suggesting that FFBN had high prediction accuracy and generalization ability.

Solar output power forecast using an ensemble framework with neural predictors and Bayesian adaptive combination

An Accurate forecast of PV output power is essential to optimize the relationship between energy supply and demand. However, it is a challenging task due to the intermittent nature of solar PV output and the effect of number of meteorological variables on it. In this paper, a multivariate neural network (NN) ensemble forecast framework is proposed. First multiple neural predictors are trained with input data from meteorological variables and then accurate predictors are combined with a Bayesian model averaging (BMA) technique. To identify the best performing framework, three different NN ensemble networks are created, namely feedforward neural network (FNN), Elman backpropagation network (ELM) and cascade-forward backpropagation (NewCF) network and trained with three different training techniques. The real time recorded solar PV data along with meteorological variables of the University of Queensland’s solar facility from 2014 to 2015 is used. To validate the forecast framework, one day ahead (24 h) forecasts are selected for different seasons. The results show that the proposed ensemble framework substantially improves the forecast accuracy of PV power output as compared benchmark methods, particularly for short term forecasting horizons.

Prediction of Partial Miscibility of Binary Mixtures Consisting of Ionic Liquids

Accurate knowledge and prediction of the liquid–liquid transition phase consisting of ionic liquids and organic solvents are key points in different industrial processes. In this way, the effects of 8 ionic liquids and 38 organic solvents through 58 binary mixtures on the miscibility gap and upper critical solution (or consolute) temperature are examined. Various approaches including 1) feedforward and cascade forward, which are trained with a backpropagation algorithm (that is, feedforward backpropagation network and cascade-forward backpropagation network) and 2) radial basis neural networks, an exact radial basis network, and generalized regression neural networks based on kernel regression, are used for the predictions and evaluations. Four independent variables (that is, molecular weight and density of ionic liquid and the organic solvent) are selected to differentiate among ionic liquids and organic solvents. In addition, the mole fraction of ionic liquid is the other input variable for the networks to predict the liquid–liquid phase transition. The obtained results reveal that the kernel regression-based networks are able to perform better as compared to two other considered networks trained with a backpropagation algorithm. The overall results indicated that the performance of the radial basis network leads to the lowest error, corresponding to the highest capability in predicting the partial miscibility condition of studied mixtures.

Artificial Neural Network Model to Predict Reservoir Storage

The rapid growth in population increases water demand thus resulting in scarcity of water which is due to improper management rather than lack of resources. Reservoir is the most important source for surface water. So, reservoir storage plays a crucial role in efficient reservoir management. Artificial neural network (ANN) is capable of simulating reservoir storage capacity. So, in present work cascade forward back propagation and feed forward back propagation network artificial neural network (ANN) models with five different architecture in each by changing the number of hidden layer neurons were developed to predict Harangi reservoir storage, Karnataka, India. The first two years (2010–2012) data on daily basis was used for supervised training and remaining (2013–2014) data on daily basis was used in prediction. The predictive accuracy using the statistical parameters like correlation coefficient (R) and mean absolute percentage error (MAPE) were found within well acceptable limit. Results shows that, compared to the cascade forward back propagation network, feed forward back propagation network architecture of 6-5-1 ANN model with correlation coefficient (R) of 0.9882 and absolute mean percentage error (MAPE) of 5.225%. is best suited for present study.

Evaluation of suspended sediment concentration using descent neural networks

This study includes three methods having considerable differences with each other and with experimental observations, because the sediment measures have certain limits. The equations relating to sediment transport are used in estimating sediment load. In the present study, black box models, ANN (Artificial Neural Network) are used for the simulation of the suspended sediment load. Hence, models which give the lowest RMSE and highest R2 are considered to be the best model for this study. The lowest values of RMSE based on normalized data for Feed forward back propagation, Cascade forward back propagation and neural network fitting are 0.00873, 0.00834 and 0.01193 respectively. The corresponding values of R2 are 0.9304, 0.9713 and 0.9831 respectively for the cited MSE. The study shows Neural Network Fitting model is superior to the other models. However, a drawback of neural network fitting is that it produces few negative estimates, which is not at all tolerable in the field of estimation of sediment load and hence these models are not crowned in this study.

SIMULTANEOUS SPECTROPHOTOMETRIC DETERMINATION OF DICLOFENAC SODIUM, PARACETAMOL, AND CHLORZOXAZONE IN TERNARY MIXTURE USING CHEMOMETRIC AND ARTIFICIAL NEURAL NETWORKS TECHNIQUES

Objective: The aim of this study is to develop and validate simple, accurate, and precise spectrophotometric methods for the simultaneous determination of diclofenac sodium (DIC), paracetamol (PAR), and chlorzoxazone (CHZ) in ternary mixture using chemometric and artificial neural networks (ANN) techniques.Methods: Three chemometric techniques include classical least squares (CLS), principal component regression (PCR), and partial least squares (PLS) in addition to cascade-forward backpropagation ANN (CFBP-ANN) were prepared using the synthetic mixtures containing the three drugs in methanol. In CLS, PCR, and PLS, the absorbances of the synthetic mixtures in the range 267-295 nm with the intervals Δλ=0.2 nm in their zero-order spectra were selected. Then, calibration or regression was obtained using the absorbance data matrix and concentration data matrix for the prediction of the unknown concentrations of DIC, PAR, and CHZ in their mixtures. In CFBP-ANN, two layers, sigmoid layer with 10 neurons and linear layer were found appropriate for the simultaneous determination of the three drugs in their ternary mixture.Results: The four proposed methods were successfully applied to the analysis of the three drugs in laboratory prepared mixtures and tablets with good percentage recoveries in the range of 98-102%. Relative standard deviation for the precision study was found <1%.Conclusion: The four proposed methods showed simplicity, accuracy, precision, and rapidity making them suitable for quality control and routine analysis of the cited drugs in ternary mixtures and pharmaceutical formulation containing them.

Multi-response optimization of ZnO thin films using Grey-Taguchi technique and development of a model using ANN

Spray pyrolysis deposited zinc oxide films have been prepared using zinc acetate precursor in iso-propyl alcohol (IPA). Based on deposition parameters, outputs were optimized for multiple performance characteristics such as thickness, roughness, and an optical transmittance of ZnO films using Grey-Taguchi technique. Optimum parameters from Grey relational analysis (GRA) were found to be 0.2M, 1bar, 365°C, and 5mLmin−1. The average thickness, roughness, and an optical transmittance obtained at the optimum condition using GRA were 79nm, 1.88nm, and 85.96% respectively as compared to the results obtained at optimized condition by a Taguchi method were 95nm, 2.45nm, and 85.19%, respectively. This indicated that the design proposed by Grey relational technique is a feasible approach for determining the optimal parameters. Further, the best condition was predicted by artificial neural network (ANN) and it was observed using cascade forward back propagation algorithm, tangent sigmoid transfer function as a hidden layer, and 4-8-3 topology.

Modeling the wire-EDM process parameters for EN-8 carbon steel using artificial neural networks

Accompanying the evolution of mechanical industry, the need for alloy materials and ceramics as well, having high toughness, hardness & impact resistance are also increasing. Wire-EDM process is a quick fix to this box. As this WEDM process is proficient in machining these materials meticulously, regardless to their toughness & hardness. The present work discusses the experimental study on wire-cut machining of EN8 Carbon Steel. This paper presents an Artificial Neural Network approach in a multi objective optimization problem, for predicting the performance measure of a machining process. The input neurons were designed by Taguchi Orthogonal Array with 5 levels each with the help of software Minitab 17. In this research work, a comparative study for predicting multiple response parameters have been carried out using Feed forward and Cascade forward back propagation algorithms of ANN. This study was conducted by varying the network designing in each case. Two different back-propagation training algorithms were used in neural network designing such as Powell-Beale Conjugate Gradient (traincgb) and Levenberg-Marquardt (trainlm). Predictive accuracy of both the models was evaluated using Mean Square Error (MSE), Root Mean Square Error (RMSE) and coefficient determination (R2). The outcomes proved that both the algorithms were able to produce feasible models. Cascade model outperformed with all statistical tests between both.Keywords: Artificial Neural Network, Feed forward back propagation, Cascade forward back propagation

Effect of Varying the Substrates and Turns of a SR in the Design of Edge Chamfered Microstrip Patch Antenna through Artificial Neural Networks and Meta Resonators

In this paper, the design of size reduced patch antenna for various wireless applications using Artificial Neural Networks (ANN) is presented. Two Network; Cascade forward Back propagation and Feed forward Back Propagation Networks are Simulated, Trained and Tested to yield the optimized design of the Conventional Patch. The conventional patch is designed to resonate at 2.785 GHz with the aid of ANN and an FEM based simulator; while the proposed rectangular patch has triple resonances 1.78, 2.58, 2.87 GHz. The antenna is conveniently modified in to a Heptagon by chamfering one of its edges. Spiral resonator is effectively used in its ground plane to create miniaturization. The Antenna has a gain of 2.368, 2.706, 4.262 dBs respectively. Keywords: Rectangular microstrip patch antenna (RMPA), Slots, Miniaturisation, Linear Polarization, WLAN, WIMAX, DCS, ANN, FF, CF.

Vibration Analysis and Prediction Natural Frequencies of Cracked Timoshenko Beam by Two Optimization Techniques - Cascade ANN and ANFIS

The prediction of the location and depth of a crack in beam type structures is an important aspect of structural health monitoring for ensuring their safety. In the present work, two methods are followed to capture the effect of open transverse crack on the fundamental as well as higher order modes of free vibration of an isotropic Timoshenko beam under clamped-clamped boundary condition using Ritz approximation. In the first method, crack beams are modelled with a rotational spring. For the second method, three-dimensional Finite Element Analysis is carried out using solid element. Dynamic characteristics are optimised by two different techniques, namely, Cascade Forward Back Propagation (CFBP) artificial neural network and Adaptive Neuro-Fuzzy Inference Systems (ANFIS). The damage is modelled and then the measured frequencies are compared to the predicted frequencies to determine the damage.

2D COORDINATE TRANSFORMATION USING ARTIFICIAL NEURAL NETWORKS

Abstract. Two coordinate systems used in Turkey, namely the ED50 (European Datum 1950) and ITRF96 (International Terrestrial Reference Frame 1996) coordinate systems. In most cases, it is necessary to conduct transformation from one coordinate system to another. The artificial neural network (ANN) is a new method for coordinate transformation. One of the biggest advantages of the ANN is that it can determine the relationship between two coordinate systems without a mathematical model. The aim of this study was to investigate the performances of three different ANN models (Feed Forward Back Propagation (FFBP), Cascade Forward Back Propagation (CFBP) and Radial Basis Function Neural Network (RBFNN)) with regard to 2D coordinate transformation. To do this, three data sets were used for the same study area, the city of Trabzon. The coordinates of data sets were measured in the ED50 and ITRF96 coordinate systems by using RTK-GPS technique. Performance of each transformation method was investigated by using the coordinate differences between the known and estimated coordinates. The results showed that the ANN algorithms can be used for 2D coordinate transformation in cases where optimum model parameters are selected.

Modelling of dust removal in rotating packed bed using artificial neural networks (ANN)

Artificial neural network (ANN) models, including the Cascade-forward back propagation neural network (CFBPNN), feed-forward back propagation neural network (FFBPNN) and Elman-forward back propagation neural network (EFBPNN), were proposed to predict the dust removal efficiency in rotating packed bed (RPB) to speed up its development. Total 326 data sets for separation grade efficiency had been collected from literatures for training and verifying the model. Gas Reynolds number (ReG), liquid Reynolds number (ReL), rotational Reynolds number (Reω), M (d02ρL/dp2/ρp) and Csi/ρG were used as input data. While, the variable η (separation grade efficiency) was taken as output data for each model. Various of hidden neurons were compared based on the mean square error (E2), coefficient of determination (R2) and residual for each model. The separation grade efficiency in RPB was also compared with other existed dust removal equipments.

Performance comparison of artificial neural network models for daily rainfall prediction

With an aim to predict rainfall one-day in advance, this paper adopted different neural network models such as feed forward back propagation neural network (BPN), cascade-forward back propagation neural network (CBPN), distributed time delay neural network (DTDNN) and nonlinear autoregressive exogenous network (NARX), and compared their forecasting capabilities. The study deals with two data sets, one containing daily rainfall, temperature and humidity data of Nilgiris and the other containing only daily rainfall data from 14 rain gauge stations located in and around Coonoor (a taluk of Nilgiris). Based on the performance analysis, NARX network outperformed all the other networks. Though there is no major difference in the performances of BPN, CBPN and DTDNN, yet BPN performed considerably well confirming its prediction capabilities. Levenberg Marquardt proved to be the most effective weight updating technique when compared to different gradient descent approaches. Sensitivity analysis was instrumental in identifying the key predictors.

Modelling of adsorption in rotating packed bed using artificial neural networks (ANN)

Rotating packed bed (RPB) exhibits good performance for absorption, distillation and extraction. However, the development of RPB was restricted by complicated liquid flow pattern on activated carbon in adsorption. Therefore, a model of artificial neural network (ANN) was adopted to predict the adsorption in RPB. The experimental data were classified into two groups, and they were training ones for establishing the model and testing ones for validating the model. Different types of ANN models, including Cascade-forward back propagation neural network (CFBPNN), Elman-forward back propagation neural network (EFBPNN) and Feed-forward back propagation neural network (FFBPNN), were investigated in this study. High gravity factor, liquid Reynolds number, adsorption time to the maximum adsorption time and packing density to liquid concentration were used as input data. While, the adsorption amount to the maximum adsorption amount was taken as output data for each model. Optimal hidden neurons for FFBPNN, EBPNN, CFBPNN were 9, 12 and 8, respectively. Experimental adsorption amount data were in good agreement with the predicted data indicating that the ANN models had a superior performance. Besides, the ANN models exhibited a more accurate prediction and had more generalization ability than multiple nonlinear regression model (MNR). The proposed FFBPNN model of absorption in RPB will provide an optimization tool to maximize the adsorption amount.