Explainable IAOA-CNN-CBAM-SVR model for predicting air consumption of auxiliary nozzles with limited sample size

Abstract

Air-jet looms are energy-intensive machines, with auxiliary nozzles accounting for nearly 80% of the total compressed air consumption. However, accurate prediction and visual analysis of nonlinear air consumption remain challenging due to limited training data and the poor interpretability of deep learning models. To address these issues, this study proposes a hybrid CNN-CBAM-SVR model optimized by an Improved Archimedes Optimization Algorithm (IAOA). Comparative experiments show that the IAOA-CNN-CBAM-SVR model achieves the lowest root mean square error (RMSE) of 0.6575, and the highest coefficient of determination (R 2 ) of 0.9941, outperforming SVR, CNN, and CNN-SVR models. Furthermore, the contributions of nozzle structural parameters to air consumption are visually illustrated using the Shapley Additive ExPlanations (SHAP) method. The findings provide a robust and interpretable model for optimizing auxiliary nozzles design and improving energy efficiency in air-jet looms.