Levenberg-Marquardt Learning Rule Research Articles

Prediction of frosting process on cold wall surface based on ANN with back propagation algorithm

The ANN with back propagation algorithm is a multi-layer feed-forward neural network, which is suitable to study unsteady frost formation with multiple factors. The back propagation ANN algorithm is used to study frost layer growth on cold flat surface, where four feature variables including temperature of cold flat surface, the velocity, relative humidity, and temperature of air are adopted. The frost growth experiment generates the database, which is good for training frost growth due to its fast speed and high precision based on Levenberg-Marquardt learning rule. The establishment of neural network model in this paper can quickly and accurately predict the frost layer height on cold flat surface of different control variables, which is helpful for the implementation of defrosting.

River flow prediction with memory-based artificial neural networks: a case study of the Dholai river basin

Prediction of hydrologic time series has been one of the most challenging tasks in water resources management due to the non-availability of adequate data. Recently, applications of artificial neural networks (ANNs) have proved quite successful in such situation in various fields. This paper demonstrates the use of memory-based ANNs to predict daily river flows. Two different networks, namely the gamma memory neural network (GMN) and genetic algorithm-gamma memory neural network (GA-GMN) have been chosen. The best network topologies for both the ANN models are achieved with Tanh transfer function and Levenberg-Marquardt learning rule after calibrations with multiple combinations of network parameters. The selected ANN models are then used to predict the daily mean flows of Dholai (Rukmi) river in Assam, India, a sub-basin of the Barak river basin. A comparative study of both networks indicates that the GA-GMN model performed better than the GMN model. The GA-GMN model gave better results for both training and testing dataset with minimum training MSE as 0.018 and minimum testing MSE as 22.97. Hence GA-GMN model is selected as an effective tool for predicting flow features of the Dholai river.

Cooperative photometric redshift estimation

AbstractIn the modern galaxy surveys photometric redshifts play a central role in a broad range of studies, from gravitational lensing and dark matter distribution to galaxy evolution. Using a dataset of ~ 25,000 galaxies from the second data release of the Kilo Degree Survey (KiDS) we obtain photometric redshifts with five different methods: (i) Random forest, (ii) Multi Layer Perceptron with Quasi Newton Algorithm, (iii) Multi Layer Perceptron with an optimization network based on the Levenberg-Marquardt learning rule, (iv) the Bayesian Photometric Redshift model (or BPZ) and (v) a classical SED template fitting procedure (Le Phare). We show how SED fitting techniques could provide useful information on the galaxy spectral type which can be used to improve the capability of machine learning methods constraining systematic errors and reduce the occurrence of catastrophic outliers. We use such classification to train specialized regression estimators, by demonstrating that such hybrid approach, involving SED fitting and machine learning in a single collaborative framework, is capable to improve the overall prediction accuracy of photometric redshifts.

Modeling of frost crystal growth over a flat plate using artificial neural networks and fractal geometries

In the present study artificial neural network and fractal geometry are used to predict frost thickness and density on a cold flat plate having constant surface temperature under forced convection for different ambient conditions. These methods are very applicable in this area because phase changes such as melting and solidification are simulated by conventional methods but frost formation is a most complicated phase change phenomenon consists of coupled heat and mass transfer. Therefore conventional mathematical techniques cannot capture the effects of all parameters on its growth and development because this process influenced by many factors and it is a time dependent process. Therefore, in this work soft computing method such as artificial neural network and fractal geometry are used to do this manner. The databases for modeling are generated from the experimental measurements. First, multilayer perceptron network is used and it is found that the back-propagation algorithm with Levenberg–Marquardt learning rule is the best choice to estimate frost growth properties due to accurate and faster training procedure. Second, fractal geometry based on the Von-Koch curve is used to model frost growth procedure especially in frost thickness and density. Comparison is performed between experimental measurements and soft computing methods. Results show that soft computing methods can be used more efficiently to determine frost properties over a flat plate. Based on the developed models, wide range of frost formation over flat plates can be determined for various conditions.

Artificial Neural Network Based Prediction of Heat Transfer From Horizontal Tube Bundles Immersed in Gas–Solid Fluidized Bed of Large Particles

Artificial neural network (ANN) modeling of heat transfer from horizontal tube bundles immersed in gas fluidized bed of large particles (mustard, raagi and bajara) was investigated. The effect of fluidizing gas velocity on the heat transfer coefficient in the immersed tube bundles in in-line and staggered arrangement is discussed. The parameters particle diameter, temperature difference between bed and immersed surface were used in the neural network (NN) modeling along with fluidizing velocity. The feed-forward network with back propagation structure implemented using Levenberg–Marquardt's learning rule in the NN approach. The predictions of the ANN were found to be in good agreement with the experiment's values, as well as the results achieved by the developed correlations.

Recovery and grade accurate prediction of pilot plant flotation column concentrate: Neural network and statistical techniques

In this study, the metallurgical performance (grade and recovery) forecasting of pilot plant flotation column using Artificial Neural Networks (ANN) and Multivariate Non-Linear Regression (MNLR) models is investigated. Modeling is performed based on 90 datasets at different operating conditions. The values of chemical reagents dosage, froth height, air and wash water flow rates, gas holdup and Cu, Mo grades in the rougher feed and flotation column feed, column tail and final concentrate streams are used to the simulation by means of NN and MNLR. The model validation analysis demonstrates the capability of both models to predict Cu and Mo grades and recoveries for a wide range of operating conditions in pilot flotation columns. It must be noted that ANN approach offers superior predictive capability over statistical method. It was also found that the error in prediction of metallurgical performance using the NN model was less than the error of the regression model. The best network is proposed with multi-layer perceptron (MLP) model, sigmoid activation function and Levenberg–Marquardt learning rule with 8-12-8-2 and 8-9-12-2 architectures, in order to estimate metallurgical performance of Cu and Mo respectively in flotation column. The results of this study indicate that a back-propagation neural network model with Root Mean Square Errors (RMSE) of 0.68 and 0.02 for prediction of Cu and Mo grades and 0.48 and 1.16 for prediction of Cu and Mo recoveries respectively has a better performance than the statistical method.

Prediction of shear strength parameters of soils using artificial neural networks and multivariate regression methods

Shear strength parameters such as friction angle and cohesion are of the most important soil's parameters which are used for design and practice in engineering works. The main aim of this paper is investigation of artificial neural networks (ANNs) and multivariate regression (MR) potential for estimation of soil shear strength parameters. For this reason, two types of ANNs including multilayer perceptron (MLP) and radial basis function (RBF), and MR including multivariate non-linear regression (MNR) as well as multivariate linear regression (MLR), have been used. Five different ANN and MR models comprising various combinations of soil's physical parameters, i.e.: percentages of passing the No. 200 (≠200), 40 (≠40) and 4 (≠4) sieves, plasticity index (PI), and density (ρ) have been developed to evaluate the effect degrees of these variables on shear strength parameters. In addition to correlation coefficient (r), root mean square error (RMSE), mean absolute error (MAE) and t-test have been also used for evaluation of prediction accuracy on both ANNs and ML methods. The results of this study indicated that MLP–ANN shows better performance rather than RBF–ANN. These results also indicated that the Levenberg–Marquardt learning rule and Sigmoid activation function were found to be appropriate for this problem. Furthermore, MLR shows better performance in prediction of shear strength parameters rather than MNR models.

Prediction of frost deposition on a horizontal circular cylinder under natural convection using artificial neural networks

In the present work Artificial Neural Network is used to predict frost thickness and density around a cooled horizontal circular cylinder having constant surface temperature under natural convection for different ambient conditions. The database for ANN generated from the experimental measurements. In the present work a multilayer perceptron network is used and it is found that the back-propagation algorithm with Levenberg–Marquardt learning rule is the best choice to estimate frost growth due to accurate and faster training procedure. Experimental measurements are used for training and testing the ANN approach and comparison is performed among the soft programming ANN and experimental measurements. It is observed that ANN soft programming code can be used more efficiently to determine frost thickness and density around a cold horizontal cylinder. Based on the developed ANN wide range of frost formation over various cylinder diameters are determined and presented for various conditions.

Development of a virtual linearizer for correcting transducer static nonlinearity

Abstract This paper reports the development of an artificial neural network based virtual linearizer for correcting nonlinearity associated with transducers connected to the data-acquisition system of a computer-based measurement system. In analog processing techniques, nonlinearity is considered to be a very serious problem that at one time was solved frequently by the piecewise linear segment approach modeled by linear electronic circuits. Since the cost of microcomputers has been reduced drastically, they are currently used in most applications of measurement, including data-acquisition subsystems. Therefore, the hardware-based analog techniques of linearization are often replaced by the software-based numerical ones. In this context, it has been found that a multilayer feed-forward back-propagation network trained with the Levenberg-Marquardt learning rule provides an optimal solution to implement an efficient soft compensator to correct transducer static-nonlinearity.

Genetic Algorithm Applied to the Selection of Conditions for the Simultaneous Quantification of Three-Food Colorants Using a Hand Scanner

A low-cost and reliable method employing a hand scanner for simultaneous colorimetric quantification of food colorant mixtures including amaranth, brilliant blue, and tartrazine is presented. Compared to a spectrophotometer, a hand scanner is inexpensive, available in most work offices, and easier to operate by non-skilled users. The appropriate instrumental conditions for measuring were selected using a genetic algorithm (GA) coupled with partial least square (PLS) regression. Using the conditions selected by GA, PLS and multiple linear regression (MLR) were compared, and similar results for the two methods were obtained. Under the selected conditions for each of the colorants, artificial neural network (ANN) including three layers of nodes and a Levenberg-Marquardt learning rule was employed, which improved the results. The concentration ranges for the three colorants in the multivariate calibration models were 0.00–5.31 mmol L−1 for amaranth, 0.00–1.85 mmol L−1 for brilliant blue, and 0.00–21.57 mmol L−1 for tartrazine. The minimum estimated relative standard error percentages (RSE%) for prediction of analytes in synthetic samples, using ANN with optimized parameters, were 16.8% for amaranth, 4.8% for brilliant blue, and 5.6% for tartrazine. A number of commercial food products were analyzed satisfactorily with the proposed method.