Optimization strategy based on genetic algorithms and neural networks applied to a polymerization process

The article presents a simple and general methodology, especially destined to the optimization of complex, strongly nonlinear systems, for which no extensive knowledge or precise models are available. The optimization problem is solved by means of a simple genetic algorithm, and the results are interpreted both from the mathematical point of view (the minimization of the objective function) and technological (the estimation of the achievement of individual objectives in multiobjective optimization). The use of a scalar objective function is supported by the fact that the genetic algorithm also computes the weights attached to the individual objectives along with the optimal values of the decision variables. The optimization strategy is accomplished in three stages: (1) the design and training of the neural model by a new method based on a genetic algorithm where information about the network is coded into the chromosomes; (2) the actual optimization based on genetic algorithms, which implies testing different values for parameters and different variants of the algorithm, computing the weights of the individual objectives and determining the optimal values for the decision variables; (3) the user's decision, who chooses a solution based on technological criteria. © 2007 Wiley Periodicals, Inc. Int J Quantum Chem, 2008     

Optimization of regenerative cycle with open feed water heater using genetic algorithms and neural networks

This article determines the operating conditions leading to maximum work in a regenerative cycle with an open feed water heater through a procedure that combines the use of artificial neural networks (ANNs) and genetic algorithms (GAs). Water is an active fluid in the thermodynamical cycle; an objective function is obtained by using vapor enthalpy (a nonlinear function of operating conditions). Utilizing classical methods for maximizing the objective function usually leads to suboptimal solutions. Therefore, this article uses ANNs to estimate the steam properties as a function of operating conditions and GAs to optimize the thermodynamical cycle. The operating conditions are chosen with the aim of gaining maximum work in a boiler for a specific heat. To estimate the thermodynamic properties, an ANN was used to provide the necessary data required in the GA calculation.     

Optimization methodology applied to feed‐forward artificial neural network parameters

This article recommends a methodology for developing a neural network with great chances to be an optimal one. The method is based on trial and error in determining the internal parameters of the network considered as having a significant influence over its performance: the number of hidden layers, activation function, number of neurons in the hidden layers, training epochs, learning rate, and momentum term. This optimization methodology is presented in two separate sections: first of them contains a series of practical considerations recommended for neural network modeling, and the second is represented by the proposed optimization algorithm, formulated in six steps and based on the practical statements. Two case studies are chosen to exemplify the use of the algorithm for finding the near optimal neural network: the dependence of the reduced and intrinsic viscosities of the siloxane‐organic copolymers of the solution concentration, temperature, and copolymer type, differing by the siloxane sequence length. The two siloxane‐organic polyazomethines resulted by the reaction of a fully aromatic bisazomethine diol with α,ω‐bis(chloromethyl)oligodimethylsiloxanes. © 2009 Wiley Periodicals, Inc. Int J Quantum Chem, 2011     

Improved neural networks based on genetic algorithm for pulse recognition

Pulse diagnosis is an important part of Chinese medicine and has played an important role in the development of Chinese medical science. However, the pulse is traditionally determined by cutting it off, which leads to a lack of objective standard pulse identification methods and affects their accuracy and feasibility. This research has studied and discussed the processing and identification of four kinds of pulse: normal pulse, wiry pulse, smooth pulse, and thready pulse. Four frequency-domain characteristics of the pulse wave and six kinds of wavelet scale energy characteristic information were extracted, and a three-layer BP (backprocessing) neural network was established. The LM (Levenberg–Marquard) algorithm and a genetic algorithm were used to improve the BP neural network, to train on and predict experimental samples, and to obtain classification accuracies of 90% and 95% respectively. Moreover, improved BP neural network based on a genetic algorithm has shown highly superior performance in terms of convergence speed and low error rate.     

Artificial neural networks applied for studying metallic complexes

Metallic complexes of multimetal and multiligand systems are complicated for calculating equilibrium concentrations in solutions. An artificial neural network has been developed for studying Al³⁺ and EDTA complexes in solution with an initial concentration of 0.01 mol L^?1 for these species. In this system there are 20 compounds and may exist 18 simultaneous reactions. The neural network has been trained and the simulated data of different concentrations as a function of pH are predicted with an accuracy of about 1% for all species simultaneously. A general analytical formula is presented, which directly relates all the concentrations as a function of pH. The analysis showed that predictions closer to the boundary of the input and output data are quantitative while out of these limits these are not even qualitative. © 2001 John Wiley & Sons, Inc. J Comput Chem 22: 1691–1701, 2001     

Artificial neural networks applied to potentiometric titration of multi-component polybasic acid mixtures

A three-layer artificial neural network model with back-propagation of error is used to treat potentiometric acid-base titration data for estimating the concentrations of individual components in polybasic weak acid mixtures. The network's architecture and parameters were optimized and an empirical rule for dynamically adjusting the learning rate is put forward to improve the network's performance. Satisfactory prediction results were obtained for three-component samples containing maleic acid, propandioic acid and succinic acid with an average relative error of 4.5%.     

Artificial neural networks based on genetic input selection for quantification in overlapped capillary electrophoresis peaks

The application of multilayer perceptron artificial neural networks (MLP ANN) based on genetic input selection for quantification of the unresolved peaks in micellar electrokinetic capillary chromatography (MECC) is reported. An optimization strategy for genetic input selection was also proposed. When the corresponding CE peaks cannot be resolved completely only by separation techniques, MLP ANN based on genetic input selection can be a suitable tool to resolve the problem. Both the spectra and the electrophoretograms of the unseparated analytes were used as the multivariate input data. The two kinds of the data were suitable for quantification of overlapped CE peaks by MLP ANN based on genetic input selection. The study also shows that the applying of genetic input selection in MLP ANN can improve the precision of quantification in both completely and partially overlapped CE peaks to some extent.     

Automated structure elucidation of organic molecules from (13)c NMR spectra using genetic algorithms and neural networks.

The automated structure elucidation of organic molecules from experimentally obtained properties is extended by an entirely new approach. A genetic algorithm is implemented that uses molecular constitution structures as individuals. With this approach, the structure of organic molecules can be optimized to meet experimental criteria, if in addition a fast and accurate method for the prediction of the used physical or chemical features is available. This is demonstrated using (13)C NMR spectrum as readily obtainable information. (13)C NMR chemical shift, intensity, and multiplicity information is available from (13)C NMR DEPT spectra. By means of artificial neural networks a fast and accurate method for calculating the (13)C NMR spectrum of the generated structures exists. The approach is limited by the size of the constitutional space that has to be searched and by the accuracy of the shift prediction for the unknown substance. The method is implemented and tested successfully for organic molecules with up to 20 non-hydrogen atoms.     

Correction of temperature variations in kinetic-based determinations by use of pruning computational neural networks in conjunction with genetic algorithms

The joint use of genetic algorithms and pruning computational neural networks is shown to be an effective means for selecting the number of inputs required to correct temperature variations in kinetic-based determinations. The genetic algorithm uses a pruning procedure based on Bayesian regularization and is highly efficient as a feature selector; it provides quite good results in the generalization process without the need to use a validation set. The fitness function is defined as the sum of two subfunctions: one controls the learning ability of the network and the other its complexity. The training, pruning, and generalization processes were initially tested with simulated data in order to acquire preliminary information for the ensuing work with real data. The performance of the proposed method was assessed by applying it to the determination of the amino acid L-glycine by its classical spectrophotometric reaction with ninhydrin. A straightforward network topology including temperature as input (40+T:2:1 with 19 connections after the pruning process) was used to estimate the L-glycine concentration from kinetic curves affected by temperature variations over the range 60-75 degrees C, using kinetic data acquired up to only 1.5 half-lives. The trained network estimates this concentration with a standard error of prediction for the testing set of ca. 8%, which is much smaller than those provided by a classical parametric method such as nonlinear regression (even if kinetic data acquired at longer half-lives are used). Finally, a kinetic interpretation of the pruning process is provided in order to better demonstrate its potential for kinetic analysis.     

Design of electroceramic materials using artificial neural networks and multiobjective evolutionary algorithms

We describe the computational design of electroceramic materials with optimal permittivity for application as electronic components. Given the difficulty of large-scale manufacture and characterization of these materials, including the theoretical prediction of their materials properties by conventional means, our approach is based on a recently established database containing composition and property information for a wide range of ceramic compounds. The electroceramic materials composition-function relationship is encapsulated by an artificial neural network which is used as one of the objectives in a multiobjective evolutionary algorithm. Evolutionary algorithms are stochastic optimization techniques which we employ to search for optimal materials based on chemical composition. The other objectives optimized include the reliability of the neural network prediction and the overall electrostatic charge of the material. The evolutionary algorithm searches for materials which simultaneously have high relative permittivity, minimum overall charge, and good prediction reliability. We find that we are able to predict a range of new electroceramic materials with varying degrees of reliability. In some cases the materials are similar to those contained in the database; in others, completely new materials are predicted.     

Forward selection radial basis function networks applied to bacterial classification based on MALDI-TOF-MS

Forward selection improved radial basis function (RBF) network was applied to bacterial classification based on the data obtained by matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF-MS). The classification of each bacterium cultured at different time was discussed and the effect of parameters of the RBF network was investigated. The new method involves forward selection to prevent overfitting and generalized cross-validation (GCV) was used as model selection criterion (MSC). The original data was compressed by using wavelet transformation to speed up the network training and reduce the number of variables of the original MS data. The data was normalized prior training and testing a network to define the area the neural network to be trained in, accelerate the training rate, and reduce the range the parameters to be selected in. The one-out-of-n method was used to split the data set of p samples into a training set of size p−1 and a test set of size 1. With the improved method, the classification correctness for the five bacteria discussed in the present paper are 87.5, 69.2, 80, 92.3, and 92.8%, respectively.     

Full-range optical pH sensor array based on neural networks

A neural network multivariate calibration is used to predict the pH of a solution in the full-range (0–14) from hue (H) values coming from imaging an optical pH sensor array based on 11 sensing elements with immobilized pH indicators. Different colorimetric acid-base indicators were tested for membrane preparation fulfilling the following conditions: 1) no leaching; 2) change in tonal coordinate by reaction and 3) covering the full pH range with overlapping between their pH responses. The sensor array was imaged after equilibration with a solution using a scanner working in transmission mode. Using software developed by us, the H coordinate of the colour space HSV was calculated from the RGB coordinates of each element.The neural network was trained with the calibration data set using the Levenberg–Marquardt training method. The network structure has 11 input neurons (each one matching the hue of a single element in the sensor array), 1 output (the pH approximation value) and 1 hidden layer with 10 hidden neurons. The network provides an MSE = 0.0098 in the training data, MSE = 0.0183 in the validation data and MSE = 0.0426 in the test data coming from a set of real water samples. The resulting correlation coefficient R obtained in the Pearson correlation test is R = 0.999.     

A toxicity evaluation and predictive system based on neural networks and wavelets

A computational approach has been developed for performing efficient and reasonably accurate toxicity evaluation and prediction. The approach is based on computational neural networks linked to modern computational chemistry and wavelet methods. In this paper, we present details of this approach and results demonstrating its accuracy and flexibility for predicting diverse biological endpoints including metabolic processes, mode of action, and hepato- and neurotoxicity. The approach also can be used for automatic processing of microarray data to predict modes of action.     

人工神经网络预测煤炭成浆浓度的研究

考虑煤炭的多种理化特性建立了成浆浓度的神经网络预测模型,对其数据预处理方法、学习率和中间层节点数等进行了深入讨论。水分、挥发分、分析基碳、灰分和氧等五个因子对于煤炭成浆性的预测起到主导作用。五因子、七因子和八因子神经网络模型对煤炭成浆浓度的预测误差分别为：0.53%、0.50%和0.74%,而现有回归分析方程的误差为1.15%,故神经网络模型比回归分析方程有更好的预测能力,尤以七因子模型最佳。     

A new optimization technique for artificial neural networks applied to prediction of force constants of large molecules

An artificial neural network (ANN) method for the prediction of force constants of chemical bonds in large, polyatomic molecules was developed. The force constant information evaluated is to be used for generating accurate estimates of the Hessian used in Newton-Raphson-type ab initio molecular structure optimization schemes. Different network topologies as well as a training procedure based on simulated annealing are evaluated. The results show that an ANN can be designed and trained to provide force constant information within a 1.5 to 5% error band even if the range of the force constants evaluated is very large (from triple bonds to hydrogen bridges). © 1995 by John Wiley & Sons, Inc.     

An accurate quantitative analysis of polymorphs based on artificial neural networks

Measurement precision based on homogeneous and accurate standard samples has been reported to result in significant improvement in the sensitivity and accuracy of the quantitative analysis of polymorphic mixtures. The purpose of this study was to further improve the accuracy of the quantitation based on data processing by artificial neural networks (ANNs), using such high quality standard samples. Homogeneous powder mixtures of - and γ-forms of indomethacin (IMC) at various ratios (0–50% -form content) were subjected to X-ray powder diffractometry. The two diffraction peaks selected as the best combination in multiple linear regression (MLR) were used in the ANN with an extended Kalman filter as a training algorithm. The results obtained by ANN had better predictive accuracy at lower contents (0–5%) compared to those of MLR. ANNs for the diffraction data based on high quality standard samples provide an extremely precise and accurate quantification for polymorphic mixtures.     

Rapid determination of radon progeny concentration based on artificial neural networks

Journal of Radioanalytical and Nuclear Chemistry - The rapid measurement of radon progeny concentration is of great significance for improving the efficiency of radon exposure dose evaluation in a...