Quantitative Analysis of Gas Phase IR Spectra Based on Extreme Learning Machine Regression Model

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Machine Learning

Background:

• Accurate quantification of components in complex gas mixtures is crucial for environmental monitoring and industrial processes.
• Traditional chemometric methods can be limited in speed and robustness when analyzing intricate spectral data.

Purpose of the Study:

• To develop and validate an advanced regression model for quantitative chemometric analysis of gas mixtures using infrared spectroscopy.
• To leverage the extreme learning machine (ELM) algorithm for improved feature extraction and regression performance.

Main Methods:

• Development of an ELM-based autoencoder (AE) for spectral signal dimensionality reduction and feature learning.
• Direct application of the ELM architecture for constructing a fast and efficient regression model.
• Analysis of simulated and experimental Fourier transform infrared spectroscopy (FTIR) data for nitrogen oxide mixtures (N2O/NO2/NO).

Main Results:

• The proposed ELM-based model demonstrated superior robustness and performance compared to conventional Principle Components Regression (PCR) and Partial Least Square Regression (PLSR) models.
• Effective dimensionality reduction and feature learning were achieved using the ELM-based autoencoder.
• Accurate quantitative analysis of nitrogen oxide mixtures was performed on both simulated and experimental FTIR data.

Conclusions:

• The developed ELM regression model offers a powerful and efficient tool for quantitative chemometric analysis of complex gas mixtures.
• This approach significantly enhances the reliability and speed of analyzing infrared spectral data.
• The model shows promise for real-world applications, such as monitoring vehicle exhaust emissions.