Hierarchical mixture of linear regressions for multivariate spectroscopic calibration

Cui, Chenhao; Fearn, Tom

doi:10.1016/j.chemolab.2017.12.013

Cited by 7 publications

(2 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Moreover, it has shown great potential in the prediction of qualitative and quantitative properties in a wide range , (e.g., agricultural products, plants, biomedicals, and pharmaceutical samples). Numerous methods, including principal component regression, multivariate linear regression, partial least squares (PLS), neural network, nonlinear PLS, and locally weighted regression, have been proposed to determine the presence of a linear or nonlinear relationship with NIR spectral data. − …”

Section: Introductionmentioning

confidence: 99%

Modified Hybrid Strategy Integrating Online Adjustable Oil Property Characterization and Data-Driven Robust Optimization under Uncertainty: Application in Gasoline Blending

et al. 2022

View full text Add to dashboard Cite

With the increasing challenges in environmental protection, product quality, and market profit, the development of a feasible optimization approach coupling accurate oil property online prediction has become crucial to the intelligent production of refineries. However, traditional property modeling with near-infrared (NIR) spectroscopy cannot dynamically extract the synergistic effect between wavelengths and cause inaccurate property prediction. The success rate of the data-driven robust optimization (DDRO) of the refining process is influenced by process uncertainty. Thus, we proposed a modified hybrid strategy on online NIR prediction modeling with adjustable characteristic wavelength selection and DDRO under uncertainty for the refining process. An adjustable feature space variable extraction (AFSVE) method was first proposed to dynamically select the characteristic wavelengths for constructing accurate property prediction models. A data-driven robust optimization of gasoline blending model was rendered through dual transformation using the uncertainty set of oil property derived from principal component analysis combined with robust kernel density estimation. An industrial application showed the best prediction accuracy of property models with the AFSVE method, and a high blending success rate was obtained by the proposed modified hybrid strategy, which hedges against uncertainties, including measuring error and blending effect, and boosts environmentally friendly and intelligent process operation.

show abstract

Section: Introductionmentioning

confidence: 99%

Modified Hybrid Strategy Integrating Online Adjustable Oil Property Characterization and Data-Driven Robust Optimization under Uncertainty: Application in Gasoline Blending

et al. 2022

View full text Add to dashboard Cite

show abstract

“…Comparison of ANN to the PLS regression has been done on various spectroscopic datasets [10]. We have also proposed some replacement methods for solving nonlinearity in multivariate regression, such as Gaussian process regression [11] and Bayesian graphical modeling [12]. Common features of these methods are high in flexibility and adaptability, but difficult to train and prone to overfitting.…”

Section: Introductionmentioning

confidence: 99%

Modern practical convolutional neural networks for multivariate regression: Applications to NIR calibration

Cui

Fearn

2018

Chemometrics and Intelligent Laboratory Systems

Self Cite

214

140

View full text Add to dashboard Cite

In this study, we investigate the use of convolutional neural networks (CNN) for near infrared (NIR) calibration. We propose a unified CNN structure that can be used for general multivariate regression purpose. The comparison between the CNN method and the partial least squares regression (PLSR) method was done on three different NIR datasets of spectra and lab reference values. Datasets are from different sources and contain 6998, 1000 and 415 training and 618, 597 and 108 validation samples, respectively. Results indicated that compared to the PLSR models, the CNN models are more accurate and less noisy. The convolutional layer in the CNN model can automatically find the suitable spectral preprocessing filter on the dataset, which significantly saves efforts in training the model.

show abstract

Basis of Matrices and Mathematical Statistics

Yun

2022

Chemometric Methods in Analytical Spectroscopy Technology

View full text Add to dashboard Cite

Hierarchical mixture of linear regressions for multivariate spectroscopic calibration

Cited by 7 publications

References 11 publications

Modified Hybrid Strategy Integrating Online Adjustable Oil Property Characterization and Data-Driven Robust Optimization under Uncertainty: Application in Gasoline Blending

Modified Hybrid Strategy Integrating Online Adjustable Oil Property Characterization and Data-Driven Robust Optimization under Uncertainty: Application in Gasoline Blending

Modern practical convolutional neural networks for multivariate regression: Applications to NIR calibration

Basis of Matrices and Mathematical Statistics

Contact Info

Product

Resources

About