Retrieval of parameters in micromixer with obstacle by cascade-forward-type artificial neural network: Comparison of four training algorithms under noisy data

Two parameters are retrieved in a passive Y-type micromixer with circular obstacle by cascade-forward-type artificial neural network (CFANN). The governing equations are solved by the finite volume method, under specific boundary conditions. The numerical model is then used to compute velocity profile and mixing efficiency, for different values of the Reynolds number. Thus, the velocity profiles along with Reynolds number (Re) and mixing efficiency (η) constitute the input–output pair of data. These data are used to train CFANN, and the network is monitored through different means, like, histograms, performance curves, and so forth. For inverse analysis, the trained CFANN model is fed with a new velocity profile as input, and corresponding values of Reynolds number and mixing efficiency are obtained as output. In an attempt to construct the optimum CFANN model, various combinations were explored, like, (1) different numbers of neurons in the hidden layer, (2) different noise levels in input data, and (3) different algorithms in the training stage. Finally, the CFANN with 10 hidden layer neurons with Levenberg–Marquardt (LM) algorithm was found to give retrieved values with up to 0.96% absolute error for all levels of noise in the input data. Also, the CFANN model with the LM algorithm has a very high value of regression coefficient of greater than 0.998, under all the noise values. Scaled conjugate gradient algorithm gives good results for the no-noise case, but fails poorly with the rise of noise. Other algorithms, like, Bayesian regularization and resilient backpropagation, perform poorly even in the no-noise case. The present approach is highly simple, accurate, and time efficient for applying inverse analysis in micromixers.     

A dynamic two-level artificial neural network for estimation of parameters in combined mode conduction-radiation heat transfer in porous medium: An application to handle huge dataset with noise

A dynamic two-level artificial neural network (DTLANN) approach is used for the estimation of parameters in combined mode conduction–radiation heat transfer in a porous medium. Four commonly used neural networks: feed forward, cascade forward, fitnet, and radial basis are used in mapping artificial neural network (ANN), and their performance is compared under noisy big data (10,302 × 1300 matrix size). Governing equations for heat transfer in the porous medium through conduction and radiation modes are solved by finite volume method and discrete transfer method. This numerical model is called a direct model. A large amount of data is generated by using the direct model for different values of extinction coefficient β and convective coupling P₂. These data were divided into different groups (class) based on the temperature difference between the gas and solid phase. In the inverse analysis, a new pair of temperature profiles for the solid and gas phase is taken as input and classified with the help of a pattern net artificial neural network model. On the basis of classification, data from that particular class and its neighbor class are used for training the mapping ANN model. After the training of the mapping ANN model, corresponding values of β and P₂ are obtained as output for any new input. This DTLANN model has a high regression coefficient (R) of .99131 and can predict highly accurate values of parameters under a huge dataset with noise, within much less CPU time.     

Modelling and prediction of bioethanol production from intermediates and byproduct of sugar beet processing using neural networks

The aim of this work was to model and predict the process of bioethanol production from intermediates and byproduct of sugar beet processing by applying artificial neural networks. Prediction of one substrate fermentation by neural networks had the same input variables (fermentation time and starting sugar content) and one output value (ethanol content, yeast cell number or sugar content). Results showed that a good prediction model could be obtained by networks with single hidden layer. The neural network configuration that gave the best prediction for raw or thin juice fermentation was one with 8 neurons in hidden layer for all observed outputs. On the other side, the optimal number of neurons in hidden layer was found to be 9 and 10 for thick juice and molasses, respectively. Further, all substrates data were merged, which led to introducing an additional input (substrate type) and defining all outputs optimal network architecture to 3-12-1. From the results the conclusion was that artificial neural networks are a good prediction tool for the selected network outputs. Also, these predictive capabilities allowed the application of the Garson's equation for estimating the contribution of selected process parameters on the defined outputs with satisfactory accuracy.     

Estimation of Energy Yield From Wind Farms Using Artificial Neural Networks

This paper uses the data from seven wind farms at Muppandal, Tamil Nadu, India, collected for three years from April 2002 to March 2005 for the estimation of energy yield from wind farms. The model is developed with the help of neural network methodology, and it involves three input variables—wind speed, relative humidity, and generation hours—and one output variable, which give the energy output from wind farms. The modeling is done using MATLAB software. The most appropriate neural network configuration after trial and error is found to be 3-5-1 (3 input layer neurons, 5 hidden layer neurons, 1 output layer neuron). The mean square error for the estimated values with respect to the measured data is $7.6 times 10^{-3}$. The results demonstrate that this work is an efficient energy yield estimation tool for wind farms.      

Energy input–output analysis and application of artificial neural networks for predicting greenhouse basil production

In this study, various Artificial Neural Networks (ANNs) were developed to estimate the production yield of greenhouse basil in Iran. For this purpose, the data collected by random method from 26 greenhouses in the region during four periods of plant cultivation in 2009–2010. The total input energy and energy ratio for basil production were 14,308,998 MJ ha^?1 and 0.02, respectively. The developed ANN was a multilayer perceptron (MLP) with seven neurons in the input layer, one, two and three hidden layer(s) of various numbers of neurons and one neuron (basil yield) in the output layer. The input energies were human labor, diesel fuel, chemical fertilizers, farm yard manure, chemicals, electricity and transportation. Results showed, the ANN model having 7-20-20-1 topology can predict the yield value with higher accuracy. So, this two hidden layer topology was selected as the best model for estimating basil production of regional greenhouses with similar conditions. For the optimal model, the values of the models outputs correlated well with actual outputs, with coefficient of determination (R²) of 0.976. For this configuration, RMSE and MAE values were 0.046 and 0.035, respectively. Sensitivity analysis revealed that chemical fertilizers are the most significant parameter in the basil production.     

Inverse neural network for optimal performance in polygeneration systems

In this paper, inverse neural network (ANNi) is applied to optimization of operating conditions or parameters in energy processes. The proposed method ANNi is a new tool which inverts the artificial neural network (ANN), and it uses a Nelder-Mead optimization method to find the optimum parameter value (or unknown parameter) for a given required condition in the process. In order to accomplish the target, first, it is necessary to build the artificial neural network (ANN) that simulates the output parameters for a polygeneration process. In general, this class of ANN model is constituted of a feedforward network with one hidden layer to simulate output layer, considering well-known input parameters of the process. Normally, a Levenberg–Marquardt learning algorithm, hyperbolic tangent sigmoid transfer-function, linear transfer-function and several neurons in the hidden layer (due to the complexity of the process) are considered in the constructed model. After that, ANN model is inverted. With a required output value and some input parameters it is possible to calculate the unknown input parameter using the Nelder-Mead algorithm. ANNi results on three different applications in energy processes showed that ANNi is in good agreement with target and calculated input data. Consequently, ANNi is applied to determine the optimal parameters, and it already has applications in different processes with a very short elapsed time (seconds). Therefore, this methodology can be useful for the controlling of engineering processes.     

Heat transfer prediction for radiant floor heating/cooling systems using artificial neural network (ANN)

Radiant floor cooling and heating systems (RHC) are gaining popularity as compared with conventional space conditioning systems. An understanding of the heat transfer capacity of the radiant system is desirable to design a space conditioning system using RHC technology. In the present work, a simplified heat flux model for RHC is developed for both cooling and heating modes of operation. The Artificial Neural Network (ANN) technique is used for the development of the simplified model. Experimental data from literature covering a wide operating range of the RHC is considered for model development and validation. Operating parameters such as mass flow rate (m_f), heat resistance (R_s), mean temperature of water flowing through the pipe (T_m), and operative temperature (T_op) are considered independent variables influencing the heat flux (q_t). The neural network consists of four input layers, one output layer, and one hidden layer with a feed-forward-back-propagation algorithm. A study on the selection of the optimum number of neurons in the range of 1–9 for the hidden layer is also performed. On the basis of the performance parameters, namely, average-absolute-relative-deviation (AARD = 0.11283) percentage, mean-square-error (MSE = 0.00055), and the coefficient of determination (R² = 0.9984), a hidden layer is modeled with five neurons.     

Artificial neural network model of molten carbonate fuel cells: Validation on experimental data

This article shows the teaching processes of artificial neural networks that are used to model the molten carbonate fuel cell (MCFC). Researchers model MCFCs to address a variety of issues across a range of complexities, from simply gauging the effect of temperature through to a complete model with 14 input parameters. The architecture of the model is a triple layer network with one hidden layer containing three neurons. The activation function used for the hidden layer was a hyperbolic tangent, with the last layer being based on linear function. We produced various network configurations, mostly networks containing one hidden layer. Models map the work of a real fuel cell with an average error in the range of 2.4% to 4.6%. The model we created guided the optimization of the thermal‐flow and construction parameters of the MCFC. Commercially available software was used to build the model and optimize the operating parameters. The selected objective functions were the efficiency of electricity production and the power density obtained from the cell's surface. The results obtained serve as pointers for possible changes in fuel cell operation and could lead to some structural changes being made.     

Estimation of crack propagation in polymer electrolyte membrane fuel cell under vibration conditions

In transportation applications, the main reasons of mechanical damage in polymer electrolyte membrane fuel cell (PEMFC) are road-induced vibrations and impact loads. The most vulnerable place of these cells is the interface between membrane and catalyst layer in the membrane electrode assembly (MEA). Hence, studies on mechanical strength of PEMFC should focus on that interface. The objective of present study lies in the fact that employing a prediction method to investigate the damage propagation behavior of vibration applied PEMFC using artificial neural network (ANN). The data available in the literature are used to constitute an ANN model. Three-layer model; input, hidden and output, are used for construction of ANN structure. Initial delamination length (a), amplitude (A), frequency (ω) and time (t) are used as input neurons whereas delamination length is output. Levenberg–Marquardt algorithm is selected as learning algorithm. On the other hand, number of hidden layer neuron is decided with the use of different neuron numbers by trial and error method. It is concluded that prediction capability of ANN model is in allowable limits and model can be suggested as efficient way of delamination length estimation.     

Modeling the effect of non-linear process parameters on the prediction of hydrogen production by steam reforming of bio-oil and glycerol using artificial neural network

Biomass-derived substrates such as bio-oil and glycerol are gaining wide acceptability as feedstocks to produce hydrogen using a steam reforming process. The wide acceptability can be attributed to a huge amount of glycerol and bio-oil obtained as by-products of biodiesel production and pyrolysis processes. Several parameters have been reported to affect the production of hydrogen by biomass steam reforming. This study investigates the effect of non-linear process parameters on the prediction of hydrogen production by biomass (bio-oil and glycerol) steam reforming using artificial neural network (ANN) modeling technique. Twenty different multilayer ANN model architectures were tested using datasets obtained from the bio-oil and glycerol steam reforming. Two algorithms namely Levenberg-Marquardt and Bayesian regularization were employed for the training of the ANNs. An optimized network configuration consisting of 3 input layer 14 hidden neurons, 1 output layer, and 3 input layer, 5 hidden neurons, and 1 output layer were obtained for the Levenberg-Marquardt and Bayesian regularization trained network, respectively for hydrogen production by bio-oil steam reforming. While an optimized network configuration consisting of 5 input nodes, 9 hidden neurons, 1 output node, and 5 input nodes, 8 hidden neurons, and 1 output node were obtained for Levenberg-Marquardt and Bayesian regularization trained network, respectively for hydrogen production by glycerol steam reforming. Based on the optimized network, the predicted hydrogen production from the bio-oil and glycerol steam agreed with the actual values with the coefficient of determination (R²) > 0.9. A low mean square error of 3.024 × 10⁻²⁴ and 6.22 × 10⁻¹⁵ for the optimized for Levenberg-Marquardt and Bayesian regularization-trained ANN, respectively. The neural network analyses of the two processes showed that reaction temperature and glycerol-to-water molar ratio were the most relevant factors that influenced the production of hydrogen by bio-oil and glycerol steam reforming, respectively. This study has demonstrated the robustness of the ANN as a technique for investigating the effect of non-linear process parameters on hydrogen production by bio-oil and glycerol steam reforming.     

Adaptive neuro‐fuzzy inference system and artificial neural network modeling of proton exchange membrane fuel cells based on nanocomposite and recast Nafion membranes

In this study, a proton exchange membrane fuel cell (PEMFC) is modeled by multilayer perceptron neural network (MLPNN), RBF neural network (RBFNN), and adaptive neuro‐fuzzy inference system (ANFIS). Experimental data are obtained on the basis of the fabricated membrane‐electrode assembly (MEA) responses using prepared nanocomposite and recast Nafion membranes in the PEMFC. Four parameters including cell temperature, inlet gas temperature, current density, and inorganic additive percent are used as inputs, and the cell voltage is considered as the output. The results show that there is no considerable discrepancy between the RBFNN accuracy (R = 0.99554) and the MLPNN accuracy (R = 0.99609) for the performance prediction. The required time for developing the RBFNN model is significantly lower than the MLPNN model. A variety of ANFIS structure is explored to approximate the behavior of the system. The effect of cell and inlet gas temperatures on the PEMFC performance is investigated by the ANFIS developed model. Predicted polarization and power–current behavior by the ANFIS for the MEA prepared by the recast Nafion and the nanocomposite membranes at the cell temperatures 50 °C to110°C are in high agreement with the experimental data. Predicted data by the ANFIS show that because of the property of Cs_2.5H_0.5PW₁₂O₄₀ additive for retaining water, much higher current density and power density at the same voltage are achieved for the nanocomposite membrane compared with the recast Nafion membrane in the PEMFC. Copyright © 2011 John Wiley & Sons, Ltd.     

Nonlinear model identification of wind turbine with a neural network

A nonlinear model of wind turbine based on a neural network (NN) is described for the estimation of wind turbine output power. The proposed nonlinear model uses the wind speed average, the standard deviation and the past output power as input data. An anemometer with a sampling rate of one second provides the wind speed data. The NN identification process uses a 10-min average speed with its standard deviation. The typical local data collected in September 2000 is used for the training, while those of October 2000 are used to validate the model. The optimal NN configuration is found to be 8-5-1 (8 inputs, 5 neurons on the hidden layer, one neuron on the output layer). The estimated mean square errors for the wind turbine output power are less than 1%. A comparison between the NN model and the stochastic model mostly used in the wind power prediction is done. This work is a basic tool to estimate wind turbine energy production from the average wind speed.     

Artificial neural network models for biomass gasification in fluidized bed gasifiers

Artificial neural networks (ANNs) have been applied for modeling biomass gasification process in fluidized bed reactors. Two architectures of ANNs models are presented; one for circulating fluidized bed gasifiers (CFB) and the other for bubbling fluidized bed gasifiers (BFB). Both models determine the producer gas composition (CO, CO₂, H₂, CH₄) and gas yield. Published experimental data from other authors has been used to train the ANNs. The obtained results show that the percentage composition of the main four gas species in producer gas (CO, CO₂, H₂, CH₄) and producer gas yield for a biomass fluidized bed gasifier can be successfully predicted by applying neural networks. ANNs models use in the input layer the biomass composition and few operating parameters, two neurons in the hidden layer and the backpropagation algorithm. The results obtained by these ANNs show high agreement with published experimental data used R² > 0.98. Furthermore a sensitivity analysis has been applied in each ANN model showing that all studied input variables are important.     

Optimum operating conditions for a water purification process integrated to a heat transformer with energy recycling using neural network inverse

Artificial neural network inverse (ANNi) is applied to calculate the optimal operating conditions on the coefficient of performance (COP) for a water purification process integrated to an absorption heat transformer with energy recycling. An artificial neural network (ANN) model is developed to predict the COP which was increased with energy recycling. This ANN model takes into account the input and output temperatures for each one of the four components (absorber, generator, evaporator, and condenser), as well as two pressures and LiBr + H₂O concentrations. For the network, a feedforward with one hidden layer, a Levenberg–Marquardt learning algorithm, a hyperbolic tangent sigmoid transfer function and a linear transfer function were used. The best fitting training data set was obtained with three neurons in the hidden layer. On the validation data set, simulations and experimental data test were in good agreement (R > 0.99). This ANN model can be used to predict the COP when the input variables (operating conditions) are well known. However, to control the COP in the system, we developed a strategy to estimate the optimal input variables when a COP is required from ANNi. An optimization method (the Nelder–Mead simplex method) is used to fit the unknown input variable resulted from the ANNi. This methodology can be applied to control on-line the performance of the system.     

Application of artificial neural networks (ANN) for modeling of industrial hydrogen plant

A back propagation feed forward artificial neural network (ANN) with three layers is used for modeling of industrial hydrogen plant. The required operating data for training of ANN is obtained by modeling and simulation of an industrial hydrogen plant. The operating data are calculated by changing effective parameters such as feed temperature, reformer pressure, steam to carbon ratio and carbon dioxide to methane ratio in feed stream. Tangent sigmoid transfer function is used in the hidden and output layer and the proposed neural network is trained with a gradient descent algorithm. The optimum number of neurons in hidden layer is determined as optimum value with minimizing of the mean square error (MSE). With changing of effective parameters, the model predicts temperature, pressure and mole fraction of hydrogen and carbon monoxide in the product of the hydrogen plant. The result can be used to gain better knowledge and optimize of the hydrogen production plants.     

Uncertainty assessment of a semi-empirical output voltage model for proton exchange membrane fuel cells

An accurate model of proton exchange membrane fuel cell (PEMFC) is essential for its characterization, performance analysis, and design of optimal control strategies. However, due to various disturbances and measurement noise in practical operation, a PEMFC presents high stochasticity and parameter uncertainty. Therefore, a semi-empirical output voltage model parameter estimation method based on variational Bayes (VB) PEMFC is proposed and combined with the Sobol sensitivity analysis method to analyze the relationship between the parameters of the model to be identified and the effect of the output voltage of the model under different noise and operating conditions. The numerical results show that the VB method is able to quantify the uncertainty of the parameter estimation results and has higher computational accuracy compared with the expectation maximization (EM) method. Compared with the Markov chain Monte Carlo (MCMC) method, the VB method is able to greatly reduce the computational effort and takes less time while satisfying the accuracy. Meanwhile, the sensitivity of the model parameters to be identified to the output voltage of the model under different noise and operating conditions is quantified using the Sobol method, which explains the variation of the posterior probability distribution results obtained using the VB method under different noise and operating conditions.     

A simple model for sizing stand alone photovoltaic systems

We consider a general model for sizing a stand-alone photovoltaic system, using as energy input data the information available in any radiation atlas. The parameters of the model are estimated by multivariate linear regression. The results obtained from a numerical sizing method were used as initial input data to fit the model. The expression proposed allows us to determine the photovoltaic array size, with a coefficient of determination ranging from 0.94 to 0.98. System parameters and mean monthly values for daily global radiation on the solar modules surface are taken as independent variables in the model. It is also shown that the proposed model can be used with the same accuracy for other locations not considered in the estimation of the model.     

Generalized neural network correlation for flow boiling heat transfer of R22 and its alternative refrigerants inside horizontal smooth tubes

The correct prediction of refrigerant boiling heat transfer performance is important for the design of evaporators. A generalized neural network correlation for boiling heat transfer coefficient of R22 and its alternative refrigerants R134a, R407C and R410A inside horizontal smooth tubes has been developed in this paper. Four kinds of dimensionless parameter groups from existing generalized correlations are selected as the input of neural network, while the Nusselt number is used as the output. Three-layer perceptron is employed as the universal approximator to build the relationship between the input and output parameters. The neuron number of hidden layer is determined by the performance of model accuracy and the standard sensitivity analysis. The experimental data of the four refrigerants in open literatures are used for correlation. The results show that the input parameter group based on the Gungor–Winterton correlation is better than the other three groups. Compared with the experimental data, the average, mean and root-mean-square deviations of the trained neural network are 2.5%, 13.0% and 20.3%, respectively, and approximately 74% of the deviations are within ±20%, which is much better than that of the existing generalized correlations.     

Daily means ambient temperature prediction using artificial neural network method: A case study of Turkey

The objective of this paper is to develop an artificial neural network (ANN) model which can be used to predict daily mean ambient temperatures in Denizli, south-western Turkey. In order to train the model, temperature values, measured by The Turkish State Meteorological Service over three years (2003–2005) were used as training data and the values of 2006 were used as testing data.In order to determine the optimal network architecture, various network architectures were designed; different training algorithms were used; the number of neuron and hidden layer and transfer functions in the hidden layer/output layer were changed. The predictions were performed by taking different number of hidden layer neurons between 3 and 30. The best result was obtained when the number of the neurons is 6. The selected ANN model of a multi-layer consists of 3 inputs, 6 hidden neurons and 1 output. Training of the network was performed by using Levenberg–Marquardt (LM) feed-forward backpropagation algorithms. A computer program was performed under Matlab 6.5 software. For each network, fraction of variance (R²) and root-mean squared error (RMSE) values were calculated and compared. The results show that the ANN approach is a reliable model for ambient temperature prediction.     

Assessment of producer gas composition in air gasification of biomass using artificial neural network model

Energy generation from renewable and carbon-neutral biomass is significant in the context of a sustainable energy framework. Hydrogen can be conveniently extracted from biomass through thermo-chemical conversion process of gasification. In the present work, an artificial neural network (ANN) model is developed using MATLAB software for gasification process simulation based on extensive data obtained from experimental investigations. Experimental investigations on air gasification are conducted in a bubbling fluidised bed gasifier with different locally available biomasses at various operating conditions to obtain the producer gas. The developed artificial neural network consists of seven input variables, output layer with four output variables and one hidden layer with fifteen neurons. The multi-layer feed-forward neural network is trained employing Levenberg–Marquardt back-propagation algorithm. Performance of the model appraised using mean squared error and regression analysis shows good agreement between the output and target values with a regression coefficient, R = 0.987 and mean squared error, MSE = 0.71. The developed model is implemented to predict the producer gas composition from selected biomasses within the operating range. This model satisfactorily predicted the effect of operating parameters on producer gas yield, and is thus a useful tool for the simulation and performance assessment of the gasification system.