基于MIV-PSO-SVM模型的矿井突水水源识别

Indentification of mine water inrush source based on MIV-PSO-SVM

为减少及防治矿井突水事故的发生,迅速、准确地判别突水水源,提出一种基于MIV(Mean Impact Value)混合粒子群优化支持向量机PSO-SVM的识别水源算法,以更有效地消除地下水源指标间的信息重叠,筛选出更好的指标体系,从而进一步提高水源识别准确率。首先,利用包含全体特征变量的水样本训练PSO-SVM网络,其次将样本分别加减一定比例构成新样本输入已训练好的网络,根据识别结果获取各影响因子的平均影响值MIV。再按照优先选取高权重变量的原则,依次剔除低权重变量,通过判断均方根误差确立最优指标体系,反馈至PSO-SVM中进行训练与预测。选取新庄孜矿实测样本进行50次试验,并与传统PSO-SVM等其他模型比较,结果表明:MIV-PSO-SVM模型可以更科学、客观地考量特征变量对预测结果的权重影响,构建更为合理的指标体系。模型预测平均准确率为94.667%,均方根误差为0.196 3,平均绝对误差百分率为3.413%,与其他模型相比,预测平均准确率明显提高,均方根误差和平均绝对误差百分率明显降低。

In order to reduce and prevent the mine water inrush accidents,and to identify the source of water inrush quickly and accurately,this paper proposes a method based on MIV（ Mean Impact Value） recognition algorithm of hybrid PSO-SVM source,to more effectively eliminate the underground water index information overlap between selected index systems better,so as to further improve the recognition accuracy of water.First,the PSO-SVM network is trained by using the water samples containing all the characteristic variables,and then the samples are added and subtracted respectively to a certain proportion to form a new sample into the trained network,according to the identification results obtaining the MIV value of each influence factor.Next,according to the principle of prior selecting high weight variables,the low weight variables are eliminated successively by judging the root mean square error to establish the optimal index system,which is fed back to the PSO-SVM for training and prediction.The selected sample is from Xinzhuangzi mine used to conduct 50 experiments,which is compared with the traditional PSO-SVM and other models.The results show that MIV-PSO-SVM model can measure the characteristic variables＇ impact on the predicted results of the weighting more scientific and objectively,to build a more reasonable index system.The average accuracy of the model is 94.667%,the root mean square error is about 0.196 3,and the mean absolute error is about 3.413%,which significantly improves the average prediction accuracy and obviously decreases the root mean square error and the mean absolute error percentage,compared with other models.