基于KPCA-APSO-ELM的矿井涌水水源识别

Identification method of mine water inrush sources based on KPCA-APSO ELM

矿井水害问题长期困扰着煤矿的安全生产,查明矿井涌水水源是矿井水害防治问题的前提。为提高矿井涌水水源的识别精度,提出了一种基于KPCA-APSO-ELM的矿井涌水水源判别模型。以袁二矿为例,在分析主要含水层地下水水化学特征的基础上,选取7种水化学离子作为判别指标。随后利用KPCA提取主要指标作为模型识别的判别因子,并通过APSO对ELM模型进行参数寻优。以63组样本数据中70%作为训练样本、30%作为预测样本进行仿真试验建立KPCA-APSO-ELM模型,并将识别结果与PCA-Logistic、KPCA-ELM和PSO-ELM模型进行比较。结果表明:KPCA算法可以有效消除指标间的冗余信息,基于KPCA-APSO-ELM模型的预测精度相对较高;与其他模型相比,该模型的均方误差和平均绝对百分比误差显著降低。

To improve recognition accuracy in discriminating the sources of mine water gushing and provide an insightful basis for mine water disaster prevention,the present paper proposes a classification method through extreme learning machine(ELM)model based on a combination of kernel principal component analysis(KPCA)and adaptive particle swarm optimization(APSO)algorithm.First,the hydro-chemical characteristics of each aquifer in the Yuandian No.2 coal mine were analyzed by mean of the Piper trilinear diagram.This analysis selected seven kinds of chemical indexes,including Ca^(2+),Mg^(2+),K^(+)+Na^(+),Cl^(-),HC0_(3)^(-),SO_(4)^(2-)and CO_(3)^(2-),labeled as identification indicators.The KPCA algorithm was subsequently used to reduce the correlation of the chemical indexes and the main control indicators were selected as discriminant factors in the identification of the water source.Moreover,APSO was equally employed to optimize the threshold and initial weight of the ELM model,resulting in the devising of a prediction model of KPCA-APSO—ELM.This model was used to carry out simulation experiments according to the 63 groups of water sample data.70%of the data were used as training samples for model learning,and 30%were used as test samples.The discriminant results of the test samples could be compared with those of the PCA-Logistic model,KPCA-ELM model,and the PSO—ELM model.The results indicate that the KPCA algorithm can effectively eliminate redundant information among the indexes.The degree of agreement between the results obtained by the KPCA-APSO-ELM and the actual situation is 94.74%.The classification accuracy of the method is remarkably higher than that of the KPCA-ELM model,PCA-Logistics model,and APSO-ELM model.Both the mean square error and the mean absolute percentage error are also significantly decreased when compared with other models.Therefore,it could be concluded that the model of KPCA-APSO-ELM generates a high prediction accuracy for the source of water gushing,which can be applied to the identification of mining inrush water sources.