矿井突水水源识别的主成分分析-混沌麻雀搜索-RF模型  

Identification of mine water inrush source based on PCA CSSA RF model

作　　者：黄敏[1] 毛岸 路世昌[1] 王彦彬[1] 邵良杉[1] HUANG Min;MAO An;LU Shichang;WANG Yanbin;SHAO Liangshan(School of Business Management,Liaoning Technical University,Huludao 125105,Liaoning,China)

机构地区：[1]辽宁工程技术大学工商管理学院,辽宁葫芦岛125000

出　　处：《安全与环境学报》2023年第8期2607-2614,共8页Journal of Safety and Environment

基　　金：国家自然科学基金项目(71771111)。

摘　　要：为快速、准确地识别矿井突水水源,根据矿井不同含水层水化学成分的差异性,将Na^(+)+K^(+)、Ca^(2+)、Mg^(2+)、Cl^(-)、SO_(4)^(2-)、HCO_(3)^(-)及总硬度作为判别指标。利用主成分分析(Principal Component Analysis,PCA)对数据进行降维,并通过混沌麻雀搜索算法(Chaotic Sparrow Search Algorithm,CSSA)对随机森林(Random Forest,RF)模型中树深和树数目参数进行寻优,建立了基于PCA-CSSA-RF的矿井突水水源识别模型。选取新庄孜矿实测的45组样本数据进行预测分析,33组数据用于模型训练,12组数据用于识别测试,并将结果与其他模型识别结果进行对比。研究表明,利用PCA对数据进行降维可以减少原始数据中的冗余,利用CSSA优化的RF模型可提高全局搜索能力和预测能力,用该模型可提高突水水源识别的效率和准确率。To quickly and accurately identify the source of water inrush,representative water chemical ions such as Na^(+)+K^(+),Ca^(2+),Mg^(2+),Cl^(-),SO_(4)^(2-),HCO_(3)^(-),and total hardness were selected as discriminatory indicators according to the variability of water chemistry in different aquifers of the mine.Based on the standardization and correlation analysis of the water sample data,the data were downscaled using Principal Component Analysis(PCA),and four principal components were extracted from the sample data as the main control indexes for water source identification.The Chaotic Sparrow Search Algorithm(CSSA)was used to find the optimum of two parameters,the tree depth and tree number,in the Random Forest(RF)algorithm,and a mine water inrush identification model based on PCA-CSSA-RF was established.Forty-five sets of sample data measured at Xinzhuangzi mine were selected for prediction analysis,33 sets of data were used for model training and 12 sets of data were used for identification testing,and the results were compared with the identification results of models such as CSSA-RF and PCA-SSA-RF.The research results show that:the PCA-CSSA-RF model has a discrimination speed of 0.9702 s and a prediction accuracy of 100%,which is faster and more accurate than other models.The iterative curve of the PCA-CSSA-RF model has the fastest convergence speed compared to other models by jumping out of the local optimal solution several times.In addition,the Mean Squared Error(MSE)and Mean Absolute Percentage Error(MAPE)are significantly lower compared to the other models.It is verified that using PCA to reduce the dimensionality of the data can effectively reduce the redundancy in the original data and reduce the complexity of the operation,and the RF model optimized by CSSA can improve the global search ability and classification efficiency.Therefore,the model can improve the efficiency and accuracy of water inrush source identification,and provide a new method for mine water inrush source identification.

关 键 词：安全工程 矿井突水 水源识别 主成分分析 混沌麻雀搜索 随机森林 

分 类 号：X936[环境科学与工程—安全科学]