Abstract
This study aims to investigate the correlation between tensile mechanical properties and chemical composition of Hippophae rhamnoides roots at different growth stages within tailings dam restoration areas, and to establish predictive models. 1-year-old, 4-year-old, and 10-year-old H. rhamnoides root systems were selected as study subjects. Through single-root tensile tests and chemical composition analysis, principal component parameters were extracted using partial least squares regression (PLSR). Combined with support vector machine (SVM) for dimensionality reduction training, a PLSR-SVM prediction model was constructed. Accuracy comparison analysis was conducted on four different prediction models.Analysis of the correlation between tensile properties and chemical composition, along with model predictions, indicates: (1) Relationship between tensile properties and age: Both the average tensile force and tensile strength of H. rhamnoides root exhibit a decreasing trend with increasing age. Regarding size effects, root tensile force follows a power-law growth pattern with increasing root diameter, while tensile strength demonstrates a power-law decay characteristic. Distinct elastic-plastic characteristics are evident in individual root stress-strain curves. Root diameter at different ages shows a negative correlation with ultimate stress, independent of strain. (2) Correlation between tensile properties and chemical composition: The tensile strength of H. rhamnoides at different ages was significantly negatively correlated with lignin content and highly significantly positively correlated with root diameter; the tensile strength was significantly positively correlated with lignin content and highly significantly negatively correlated with root diameter. The tensile properties of H. rhamnoides at different ages were significantly different from each component of the root system, and the single chemical component content could not fully explain the size effect between root diameter and tensile properties. (3) Predictive performance of models linking tensile properties to chemical composition: The PLSR-SVM model demonstrated optimal overall performance in predicting tensile properties, with the evaluated error metrics significantly outperforming traditional the principal component analysis (PCA) model, the back propagation neural network (BP) model, the PLSR model, and the SVM model.