Get started
Work

An evaluation of improved gradient boosting models for predicting dynamic normal stress on silo wall

Year: 2025
Type: article
Abstract: ABSTRACT An improved gradient boosting model is proposed based on intelligent optimization algorithms to predict dynamic normal stress on silo walls assisted with 874 datasets with 10 influencing factors. Highly correlated variables were screened through Pearson correlation coefficients to reduce dimensionality. To address the multi-peak stress distribution of dynamic normal stress caused by particle accumulation at the bottom of silo, we developed a hybrid model combining Whale Optimization Algorithm (WOA) and Sparrow Search Algorithm (SSA) with XGBoost and LightGBM to solve high-dimensional nonlinear problems. WOA employs its unique encirclement-spiral dual-mode optimization mechanism to efficiently tune the hyperparameters of both XGBoost and LightGBM through adaptive weight adjustment. The fitness values stabilize within fewer iterations across different population sizes, significantly improving the computational efficiency of hyperparameter optimization for the hybrid model. SSA explores hyperparameter combinations from diverse perspectives by simulating sparrows' foraging and anti-predation behaviors, effectively preventing convergence to local optima. The results show that compared with the fitting ability and generalization ability of other single models, the performance of XGboost model is significantly improved, and its relative error is small. Furthermore, SHAP values were employed to quantify feature contributions to the model predictions. The dimension of discharge port for silo structure and the coefficient of internal friction angle for granular materials should be priority consideration when to design the silo structure.
DOI: 10.1016/j.rineng.2025.106961
Language: English
Open Access status: gold
Estimated APC paid: $1,940
OpenAlex ID: W4413825118
FWCI: 0.427
Cites: 40
Cited by: 1
Related to: 10