Scholarlink

<p>Mine backfilling, as a mining environmental protection and safety technology that simultaneously addresses goaf support and tailings disposal, has been widely applied in the management of goafs after underground mining operations. Cemented tailings backfill materials are of great significance in underground backfilling due to their ability to effectively control ground pressure, low cost, and contribution to reducing surface tailings accumulation. However, the traditional method of determining the strength of cemented tailings backfill materials through laboratory compressive tests to meet the design requirements of mining plans is both time-consuming and labor-intensive. To address the inefficiency in traditional experimental methods, this study establishing a dataset containing 745 sets of data, combined Bayesian Optimization (BO) with four single machine learning models, namely Gaussian Process Regression (GPR), Linear Regression (LR), Stacked-ensemble, and Local Cascade Ensemble (LCE), developing four hybrid machine learning models (BO-GPR, BO-LR, BO-Stacked-ensemble, and BO-LCE) primarily for predicting the unconfined compressive strength (UCS) of 32.5-grade ordinary Portland cemented tailings backfill materials. After comparing the fitting effects of scatter plots, the distribution of residual plots, and evaluating indicators, the study found that BO-GPR is the optimal model, with the highest coefficient of determination (R<sup>2</sup>) and the lowest root mean square error (RMSE). Through Shapley Additive exPlanations (SHAP) analysis, it was found that the cement-tailings ratio (CTR) is the most critical factor affecting the UCS, followed by curvature coefficient (Cc), temperature (TEM), mass concentration (MC), Al₂O₃ content, SiO₂ content, CaO content, curing time (T), uniformity coefficient (Cu), and particle size distribution parameter D<sub>50</sub>. The BO-GPR model proposed in this study can be used to predict the UCS of cemented tailings backfill materials, and provide preliminary technical guidance for the optimization of on-site backfill engineering.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Interpretable Machine Learning for Predicting Unconfined Compressive Strength of Cemented Tailings Backfill: A Data-Driven Approach with Bayesian Optimization and SHAP Analysis

  • 期刊论文

  • 发表日期：

    2026年7月11日

查看全文
  • Hongwei Wang,
  • Hui Ye,
  • Longjun Dong,
  • Daoyuan Sun

摘要

Mine backfilling, as a mining environmental protection and safety technology that simultaneously addresses goaf support and tailings disposal, has been widely applied in the management of goafs after underground mining operations. Cemented tailings backfill materials are of great significance in underground backfilling due to their ability to effectively control ground pressure, low cost, and contribution to reducing surface tailings accumulation. However, the traditional method of determining the strength of cemented tailings backfill materials through laboratory compressive tests to meet the design requirements of mining plans is both time-consuming and labor-intensive. To address the inefficiency in traditional experimental methods, this study establishing a dataset containing 745 sets of data, combined Bayesian Optimization (BO) with four single machine learning models, namely Gaussian Process Regression (GPR), Linear Regression (LR), Stacked-ensemble, and Local Cascade Ensemble (LCE), developing four hybrid machine learning models (BO-GPR, BO-LR, BO-Stacked-ensemble, and BO-LCE) primarily for predicting the unconfined compressive strength (UCS) of 32.5-grade ordinary Portland cemented tailings backfill materials. After comparing the fitting effects of scatter plots, the distribution of residual plots, and evaluating indicators, the study found that BO-GPR is the optimal model, with the highest coefficient of determination (R2) and the lowest root mean square error (RMSE). Through Shapley Additive exPlanations (SHAP) analysis, it was found that the cement-tailings ratio (CTR) is the most critical factor affecting the UCS, followed by curvature coefficient (Cc), temperature (TEM), mass concentration (MC), Al₂O₃ content, SiO₂ content, CaO content, curing time (T), uniformity coefficient (Cu), and particle size distribution parameter D50. The BO-GPR model proposed in this study can be used to predict the UCS of cemented tailings backfill materials, and provide preliminary technical guidance for the optimization of on-site backfill engineering.