Factor Screening for Equivalence Ratio Measurement Modeling: A Hybrid Approach Integrating Random Forest Algorithm and Support Vector Regression

Measuring the equivalence ratio using flame spectral data is a key focus in combustion diagnostic techniques. Traditional methods rely on chemiluminescent bands with distinct spectral features, which may not fully leverage the richness of hyperspectral data. In this article, we propose a method for screening strongly correlated spectral features for equivalence ratio measurement modeling from spectral data using the random forest (RF) algorithm. The RF evaluates and ranks the influence of each wavelength on the equivalence ratio regression model by calculating the mean square error (mse) at each feature wavelength. Wavelengths with higher scores are selected as model factors, which are then used to construct specific equivalence ratio measurement models by support vector regression (SVR). The constructed equivalence ratio measurement model achieved an mse of 0.0019 and a coefficient of determination ( $R^{2}$ ) value of 0.96, demonstrating the model’s strong predictive capability and generalization ability.