Adaptive modeling method integrating slime mould algorithm and cascade ensemble: Nondestructive detection of silage quality under VIS-NIRS

Abstract

Rapid and scientific evaluation of silage quality is essential for livestock farming. The aim is fast, large-scale, and non-destructive detection of silage pH and quality grades. The Slime Mould Algorithm (SMA) was integrated with a cascade ensemble (cascading) to create an intelligent and adaptive modeling algorithm (SMA-configured cascading). Firstly, visible-near-infrared spectra of aerobically deteriorated silage were collected and preprocessed. Secondly, SMA was employed to mine spectral features. Finally, SMA-configured cascading was applied for adaptive modeling by configuring the learners. The results demonstrated that 39 features extracted by SMA performed optimally regarding predictive effectiveness compared to two benchmark algorithms. These features effectively captured key quality information whereas avoiding interference. The SMA-configured cascading achieved the best prediction accuracy for silage quality, outperforming conventional adaptive-based and single-learner-based modeling methods. For pH prediction, the $R_p^2$, RMSE_P, MAE_P, MAPE_P, RPD, configuration time (ET_con), and prediction time (ET_pre) of the prediction set were 0.9954, 0.1020, 0.0750, 1.6836 %, 14.9808, 19070 s, and 20.98 s, respectively. The optimal cascading configuration was Partial Least Squares Regression (PLSR), PLSR, support vector machine (SVM), and SVM. For quality grade determination, the Accuracy_p, F1-score_p, ET_con, and ET_pre of the prediction set were 86.11 %, 0.8639, 3097.47 s, and 21.49 s, respectively, with the configured cascading being adaptive boosting and K-nearest neighbor. The proposed method enables efficient and adaptive modeling based on spectral features, optimizing quality prediction. It holds the potential for in-situ detection through offline configuration and online prediction. This study provides theoretical and technical support for the rapid assessment of silage quality in production environments.