J Shanghai Jiaotong Univ Sci ›› 2023, Vol. 28 ›› Issue (5): 686-694.doi: 10.1007/s12204-022-2408-7

• Mining Engineering • Previous Articles    

Prediction of Backfill Strength Based on Support Vector Regression Improved by Grey Wolf Optimization

基于灰狼优化算法改进支持向量回归的充填体强度预测研究

ZHANG Bo(张博),LI Keqing (李克庆),HU Yafei(胡亚飞),JI Kun(吉坤),HAN Bin*(韩斌)   

  1. (School of Civil and Resource Engineering, University of Science and Technology Beijing, Beijing 100083, China)
  2. (北京科技大学 土木与资源工程学院,北京 100083)
  • Accepted:2021-09-09 Online:2023-09-28 Published:2023-10-20

Abstract: In order to predict backfill strength rapidly with high accuracy and provide a new technical support for digitization and intelligentization of mine, a support vector regression (SVR) model improved by grey wolf optimization (GWO), GWO-SVR model, is established. First, GWO is used to optimize penalty term and kernel function parameter in SVR model with high accuracy based on the experimental data of uniaxial compressive strength of filling body. Subsequently, a prediction model which uses the best two parameters of best c and best g is established with the slurry density, cement dosage, ratio of artificial aggregate to tailings, and curing time taken as input factors, and uniaxial compressive strength of backfill as the output factor. The root mean square error of this GWO-SVR model in predicting backfill strength is 0.143 and the coefficient of determination is 0.983, which means that the predictive effect of this model is accurate and reliable. Compared with the original SVR model without the optimization of GWO and particle swam optimization (PSO)-SVR model, the performance of GWO-SVR model is greatly promoted. The establishment of GWO-SVR model provides a new tool for predicting backfill strength scientifically.

Key words: underground mining, backfill strength, prediction model, grey wolf optimization (GWO), support vector regression (SVR)

摘要: 为了高精度快速预测矿山充填体强度并为矿山数字化、智能化提供新技术支持,本文建立基于灰狼优化算法改进的支持向量回归(GWO-SVR)模型。首先,基于充填体单轴抗压强度实验数据,利用灰狼优化算法(GWO)对支持向量回归(SVR)中的惩罚系数c和核函数参数g进行高精度优化。然后以料浆浓度、水泥掺量、人工砂尾砂比和养护龄期为输入变量,以充填体单轴抗压强度为输出变量,建立了使用最佳两个最优参数best_c 和 best_g的预测模型。该GWO-SVR模型预测充填体强度的均方根误差为 0.143,可决系数为 0.983,说明该模型的预测效果准确可靠。与未经 GWO 优化的原始 SVR 模型和 PSO-SVR 模型相比,GWO-SVR 模型的性能得到了极大的提升。GWO-SVR 模型的建立为科学预测矿山充填体强度提供了新的工具。

关键词: 地下采矿,充填体强度,预测模型,灰狼优化算法,支持向量回归

CLC Number: