Prediction- and simulation-error based perceptron training: Solution space analysis and a novel combined training scheme

Previous papers have noted the difficulty in obtaining neural models which are stable under simulation when trained using prediction-error-based methods. Here the differences between series-parallel and parallel identification structures for training neural models are investigated. The effect of the error surface shape on training convergence and simulation performance is analysed using a standard algorithm operating in both training modes. A combined series-parallel/parallel training scheme is proposed, aiming to provide a more effective means of obtaining accurate neural simulation models. Simulation examples show the combined scheme is advantageous in circumstances where the solution space is known or suspected to be complex. (c) 2006 Elsevier B.V. All rights reserved.

A new battery modelling method based on simulation error minimization

A new battery modelling method is presented based on the simulation error minimization criterion rather than the conventional prediction error criterion. A new integrated optimization method to optimize the model parameters is proposed. This new method is validated on a set of Li ion battery test data, and the results confirm the advantages of the proposed method in terms of the model generalization performance and long-term prediction accuracy.

Rapid, quantitative analysis of ppm/ppb nicotine using surface-enhanced Raman scattering from polymer-encapsulated Ag nanoparticles (gel-colls)

Rapid, quantitative SERS analysis of nicotine at ppm/ppb levels has been carried out using stable and inexpensive polymer-encapsulated Ag nanoparticles (gel-colls). The strongest nicotine band (1030 cm(-1)) was measured against d(5)-pyridine internal standard (974 cm(-1)) which was introduced during preparation of the stock gel-colls. Calibration plots of I-nic/I-pyr against the concentration of nicotine were non-linear but plotting I-nic/I-pyr against [nicotine](x) (x = 0.6-0.75, depending on the exact experimental conditions) gave linear calibrations over the range (0.1-10 ppm) with R-2 typically ca. 0.998. The RMS prediction error was found to be 0.10 ppm when the gel-colls were used for quantitative determination of unknown nicotine samples in 1-5 ppm level. The main advantages of the method are that the gel-colls constitute a highly stable and reproducible SERS medium that allows high throughput (50 sample h(-1)) measurements.

Locally regularised two-stage learning algorithm for RBF network centre selection

Nonlinear models constructed from radial basis function (RBF) networks can easily be over-fitted due to the noise on the data. While information criteria, such as the final prediction error (FPE), can provide a trade-off between training error and network complexity, the tunable parameters that penalise a large size of network model are hard to determine and are usually network dependent. This article introduces a new locally regularised, two-stage stepwise construction algorithm for RBF networks. The main objective is to produce a parsomous network that generalises well over unseen data. This is achieved by utilising Bayesian learning within a two-stage stepwise construction procedure to penalise centres that are mainly interpreted by the noise.

Semi batch learning with store management using enhanced conjugate gradient

This paper explores the performance of sliding-window based training, termed as semi batch, using multilayer perceptron (MLP) neural network in the presence of correlated data. The sliding window training is a form of higher order instantaneous learning strategy without the need of covariance matrix, usually employed for modeling and tracking purposes. Sliding-window framework is implemented to combine the robustness of offline learning algorithms with the ability to track online the underlying process of a function. This paper adopted sliding window training with recent advances in conjugate gradient direction with application of data store management e.g. simple distance measure, angle evaluation and the novel prediction error test. The simulation results show the best convergence performance is gained by using store management techniques. © 2012 Springer-Verlag.

Modeling of titanium alloys by using artificial neural networks

Titanium alloy exhibits an excellent combination of bio-compatibility, corrosion resistance, strength and toughness. The microstructure of an alloy influences the properties. The microstructures depend mainly on alloying elements, method of production, mechanical, and thermal treatments. The relationships between these variables and final properties of the alloy are complex, non-linear in nature, which is the biggest hurdle in developing proper correlations between them by conventional methods. So, we developed artificial neural networks (ANN) models for solving these complex phenomena in titanium alloys.

In the present work, ANN models were used for the analysis and prediction of the correlation between the process parameters, the alloying elements, microstructural features, beta transus temperature and mechanical properties in titanium alloys. Sensitivity analysis of trained neural network models were studied which resulted a better understanding of relationships between inputs and outputs. The model predictions and the analysis are well in agreement with the experimental results. The simulation results show that the average output-prediction error by models are less than 5% of the prediction range in more than 95% of the cases, which is quite acceptable for all metallurgical purposes.

Prediction of adipose tissue composition using Raman spectroscopy: average properties and individual fatty acids.

Abstract: Raman spectroscopy has been used for the first time to predict the FA composition of unextracted adipose tissue of pork, beef, lamb, and chicken. It was found that the bulk unsaturation parameters could be predicted successfully [R-2 = 0.97, root mean square error of prediction (RMSEP) = 4.6% of 4 sigma], with cis unsaturation, which accounted for the majority of the unsaturation, giving similar correlations. The combined abundance of all measured PUFA (>= 2 double bonds per chain) was also well predicted with R-2 = 0.97 and RMSEP = 4.0% of 4 sigma. Trans unsaturation was not as well modeled (R-2 = 0.52, RMSEP = 18% of 4 sigma); this reduced prediction ability can be attributed to the low levels of trans FA found in adipose tissue (0.035 times the cis unsaturation level). For the individual FA, the average partial least squares (PLS) regression coefficient of the 18 most abundant FA (relative abundances ranging from 0.1 to 38.6% of the total FA content) was R-2 = 0.73; the average RMSEP = 11.9% of 4 sigma. Regression coefficients and prediction errors for the five most abundant FA were all better than the average value (in some cases as low as RMSEP = 4.7% of 4 sigma). Cross-correlation between the abundances of the minor FA and more abundant acids could be determined by principal component analysis methods, and the resulting groups of correlated compounds were also well-predicted using PLS. The accuracy of the prediction of individual FA was at least as good as other spectroscopic methods, and the extremely straightforward sampling method meant that very rapid analysis of samples at ambient temperature was easily achieved. This work shows that Raman profiling of hundreds of samples per day is easily achievable with an automated sampling system.

Preliminary investigation of the application of Raman spectroscopy to the prediction of the sensory quality of beef silverside

The potential of Raman spectroscopy for the determination of meat quality attributes has been investigated using data from a set of 52 cooked beef samples, which were rated by trained taste panels. The Raman spectra, shear force and cooking loss were measured and PLS used to correlate the attributes with the Raman data. Good correlations and standard errors of prediction were found when the Raman data were used to predict the panels' rating of acceptability of texture (R-2 = 0.71, Residual Mean Standard Error of Prediction (RMSEP)% of the mean (mu) = 15%), degree of tenderness (R-2 = 0.65, RMSEP% of mu = 18%), degree of juiciness (R-2 = 0.62, RMSEP% of mu = 16%), and overall acceptability (R-2 = 0.67, RMSEP% of mu = 11%). In contrast, the mechanically determined shear force was poorly correlated with tenderness (R-2 = 0.15). Tentative interpretation of the plots of the regression coefficients suggests that the alpha-helix to beta-sheet ratio of the proteins and the hydrophobicity of the myofibrillar environment are important factors contributing to the shear force, tenderness, texture and overall acceptability of the beef. In summary, this work demonstrates that Raman spectroscopy can be used to predict consumer-perceived beef quality. In part, this overall success is due to the fact that the Raman method predicts texture and tenderness, which are the predominant factors in determining overall acceptability in the Western world. Nonetheless, it is clear that Raman spectroscopy has considerable potential as a method for non-destructive and rapid determination of beef quality parameters.

Thermogravimetric evaluation of perennial ryegrass (Lolium perenne) for the prediction of in vitro dry matter digestibility

Thermogravimetry (TG) can be used for assessing the compositional differences in grasses that relate to dry matter digestibility (DMD) determined by pepsin-cellulase assay. This investigation developed regression models for predicting DMD of herbage grass during one growing season using TG results. The calibration samples were obtained from a field trial of eight cultivars and two breeding lines. The harvested materials from five cuts were analysed by TG to identify differences in the combustion patterns within the range of 30-600 degrees C. The discrete results including weight loss, peak height, area, temperature, widths and residue of three decomposition peaks were regressed against the measured DMD values of the calibration samples. Similarly, continuous weight loss results of the same samples were also utilised to generate DMD models. The r(2) for validation of the discrete and the best continuous models were 0.90 and 0.95, respectively, and the two calibrations were validated using independent samples from 24 plots from a trial carried out in 2004. The standard error for prediction of the 24 samples by the discrete model (4.14%) was higher than that by the continuous model (2.98%). This study has shown that DMD of grass could be predicted from the TG results. The benefit of thermal analysis is the ability to detect and show changes in composition of cell wall fractions of grasses during different cuts in a year.

The prediction of intake potential and organic matter digestibility of grass silages by near infrared spectroscopy analysis of undried samples

A study was undertaken to examine a range of sample preparation and near infrared reflectance spectroscopy (NIPS) methodologies, using undried samples, for predicting organic matter digestibility (OMD g kg(-1)) and ad libitum intake (g kg(-1) W-0.75) of grass silages. A total of eight sample preparation/NIRS scanning methods were examined involving three extents of silage comminution, two liquid extracts and scanning via either external probe (1100-2200 nm) or internal cell (1100-2500 nm). The spectral data (log 1/R) for each of the eight methods were examined by three regression techniques each with a range of data transformations. The 136 silages used in the study were obtained from farms across Northern Ireland, over a two year period, and had in vivo OMD (sheep) and ad libitum intake (cattle) determined under uniform conditions. In the comparisons of the eight sample preparation/scanning methods, and the differing mathematical treatments of the spectral data, the sample population was divided into calibration (n = 91) and validation (n = 45) sets. The standard error of performance (SEP) on the validation set was used in comparisons of prediction accuracy. Across all 8 sample preparation/scanning methods, the modified partial least squares (MPLS) technique, generally minimized SEP's for both OMD and intake. The accuracy of prediction also increased with degree of comminution of the forage and with scanning by internal cell rather than external probe. The system providing the lowest SEP used the MPLS regression technique on spectra from the finely milled material scanned through the internal cell. This resulted in SEP and R-2 (variance accounted for in validation set) values of 24 (g/kg OM) and 0.88 (OMD) and 5.37 (g/kg W-0.75) and 0.77 (intake) respectively. These data indicate that with appropriate techniques NIRS scanning of undried samples of grass silage can produce predictions of intake and digestibility with accuracies similar to those achieved previously using NIRS with dried samples. (C) 1998 Elsevier Science B.V.