Estudo comparativo entre tecnicas de inteligencia artificial e modelos lineares em determinações quantitativas no infravermelho proximo

AUTOR(ES)

Paulo Augusto da Costa Filho

DATA DE PUBLICAÇÃO

2003

RESUMO

In this work it was done a comparative study among the performances of different artificial intelligence techniques and two multivariate linear calibration methods (Multiple Linear Regression and Partial Least Squares), in three different chemical systems for quantitative determination in the near infrared region. The artificial intelligence techniques are a particular case of a non-linear models, which use intelligent algorithms to describe the behavior of one system in an iterative way. The artificial intelligence techniques studied were: genetic algorithm, neural network and Fuzzy logic, as well as hibrid systems as neurogenetic and Neuro-Fuzzy. This thesis shows the results obtained from three different chemical systems: 1- solutions of sugar cane juice; 2- a mix between powders of a polymer and one special mineral; 3- determination of glucose, cholesterol and triglycerides in human plasma. It was established a correlation betwen near infrared spectrum and Pol and Brix degrees, both parameters are used to evaluate the sugar quality. For this analysis, the best calibration model gave a standard error of prediction for Brix and Pol of 0.35 and 0.55, respectively. For the second system, it was used the diffuse reflectance in the near infrared range to analyse the mineral content in a polymer. The aim of this analysis was to determine the mineral concentration in this solid mix because it provides special chemical and physical properties to the polymer. The near infrared analyses provided very good results: it was obtained a standard error of prediction of 0.009%. To conclude this work, it was done quantitative analysis of glucose, cholesterol and triglycerides in human plasma. For glucose, total cholesterol and triglycerides it was obtained standard errors of prediction of 6.54, 13.39 and 14.73 mg dL, respectively.

ASSUNTO(S)

inteligencia artificial quimiometria espectroscopia de infravermelho

Documentos Relacionados

• Redes neurais e tecnicas de pre-processamento de sinais em espectroscopia de reflectancia no infravermelho proximo
• Aplicação de técnicas multivariadas e inteligência artificial na análise de espectros de infravermelho para determinação de matéria orgânica em amostras de solos
• Estudo comparativo entre modelos lineares e redes neurais artificiais como tecnologias geradoras de previsões de valores financeiros
• Desenvolvimento e aplicação de modelos de transferencia de calibração em espectroscopia no infravermelho proximo de ondas curtas (SW-NIR)
• Workflow com técnicas de planejamento apoiado em inteligência artificial