基于机器学习算法及Stacking融合集成模型的矿柱稳定性分析  

Ore Pillar Stability Analysis Based on Machine Learning Algorithms and Stacking Fusion Ensemble Models

作　　者：张文革 董陇军[2] 王加闯 龚甦文 罗才严 郝晨良 曹恒 ZHANG Wenge;DONG Longjun;WANG Jiachuang;GONG Suwen;LUO Caiyan;HAO Chenliang;CAO Heng(Sifang Gold Mine Co.,Ltd.,Baoji 721000,China;School of Resources and Safety Engineering,Central South University,Changsha 410083,China)

机构地区：[1]陕西凤县四方金矿有限责任公司,陕西宝鸡721000 [2]中南大学资源与安全工程学院,湖南长沙410083

出　　处：《金属矿山》2023年第10期67-74,共8页Metal Mine

基　　金：“十四五”国家重点研发计划项目(青年科学家项目)(编号:2021YFC2900500)。

摘　　要：留设矿柱作为确保矿山地下安全开采的重要手段,开展其稳定性研究对矿山的安全生产具有重要意义。为此,基于机器学习算法及Stacking融合策略开展了矿柱稳定性分析。首先,通过对原始矿柱稳定性数据样本进行统计分析,利用SMOTE(Synthetic Minority Over-sampling Technique)算法对原始数据进行了样本平衡化处理,并按照80%的数据作为训练集、20%的数据作为测试集进行划分。其次,使用随机森林算法(Random Forest,RF)、K-近邻算法(K-nearest Neighbor,KNN)、支持向量机算法(Support Vector Machine,SVM)、线性判别降维算法(Linear Discriminant Dimensionality Reduction,LDA)、多层神经网络算法(Multi-layer Neural Network,MLPC)以及逻辑回归算法(Logistic Regression,LR)等不同算法进行分类计算。然后,通过随机搜索算法和五折交叉验证来获取每个模型的最优超参数,并分别选取上述单个方法为元模型,结合Stacking融合策略构建6种集成模型。最后,通过对比评价模型的准确率、召回率、精确率和F_(1)指数等指标来确定最佳的评估方法。研究表明:在传统机器学习算法中,SVM算法在分类任务中表现最优,而在采用Stacking融合策略的集成模型中,以随机森林作为元模型的Stacking模型展现出最佳性能;此外,通过采用Stacking融合策略,整个集成算法模型相较于各个算法对应的元模型,性能得到明显提升。As an important means to ensure the safety of underground mining,the research on the stability of pillar is of great significance to the safety production of mine.To this end,pillar stability analysis is carried out based on machine learning algorithm and Stacking fusion strategy.Firstly,through statistical analysis of the original pillar stability data samples,SMOTE(Synthetic Minority Over sampling Technique)algorithm was used to balance the original data,and 80%of the data was used as the training set,20%of the data is divided as a test set.Secondly,Random Forest(RF)algorithm,K-nearest Neighbor(KNN)algorithm,Support Vector Machine(SVM)algorithm and Linear Discriminant Reduction Algorithm(LDA),Multilayer Neural Network(MLPC)and Logistic Regression(LR)algorithm are used for classification calculation.Then,the optimal hyperparameters of each model are obtained by random search algorithm and 50-fold cross-validation,and the above methods are selected as the meta-model respectively.Six integrated models are constructed in combination with the Stacking fusion strategy.Finally,the best evaluation method is determined by comparing the accuracy rate,recall rate,accuracy rate and F_(1) index of the evaluation model.The research shows that:in the traditional machine learning algorithms,SVM algorithm performs the best in classification tasks,and in the integrated models that adopt the Stacking strategy,the stochastic forest as the metamodel shows the best performance.In addition,by adopting the Stacking fusion strategy,the entire integrated algorithm model has significant performance improvement compared with the metametmodels corresponding to each algorithm.

关 键 词：矿柱稳定性 超参数优化 Stacking融合策略 性能评估 

分 类 号：TD853[矿业工程—金属矿开采]