基于PCA-KD-KNN方法的矿井突水水源判别分析研究

Study on Discriminant Analysis of Mine Water Inrush Source Based on PCA-KD-KNN Method

在煤矿开采过程中,矿井突水事故严重威胁着煤矿安全生产和工人的生命安全。为了快速准确地判别矿井突水水源,达到有效预防水害事故的目的,基于KD-tree(K-dimension tree)与KNN(K-Nearest Neighbor algorithm,KNN)算法,建立了矿井突水水源判别方法。根据矿井中不同含水层的水化学特征的差异性,选取9种水化学成分作为突水水源的判别指标。采用主成分分析法(PCA)进行数据降维;进一步运用K维树形结构存储训练样本,提高数据搜索效率,然后结合KNN算法进行突水水源判别。以蔚州矿区为例,采用矿区4个含水层的24组实测数据构建模型,其中16组作为训练样本,另外8组为测试样本,并与传统KNN算法的判别结果进行对比。结果表明:KD-tree确定了离待测样本最邻近的3个训练样本,降低了KNN算法的计算复杂度。对比KD-tree与KNN相结合的新方法与传统KNN算法的判别结果,新方法的准确率提高了25%,说明新方法能使判别结果更加快速准确。

In the process of coal mining,mine water inrush accident is a serious threat to the production safety of coal mine and workers’life.In order to quickly and accurately identify the mine water inrush source,and to achieve the purpose of effectively preventing water disasters,the mine water inrush source identification method was established based on KD-tree(K-dimension tree)and KNN(K-Nearest Neighbor algorithm,KNN).According to the differences in hydrochemical characteristics of different aquifers in the mine,nine water chemical components were selected as the discriminant indexes of water inrush sources.Principal component analysis(PCA)was used for data dimension reduction,the K-dimensional tree structure was further used to store the training samples to improve the data search efficiency,and then KNN algorithm was used to identify the water inrush source.Taking the Yuzhou mining area as an example,24 groups of measured data of 4 aquifers in the mining area were used to construct the model,among which 16 groups were used as training samples,and the other 8 groups were used as test samples.The discriminant results were compared with those of the traditional KNN algorithm.The results show that KD-tree determines the three nearest training samples to the test sample,which reduces the computational complexity of KNN algorithm.Comparing the discriminant results of the new method combining KD-tree and KNN with the traditional KNN algorithm,the accuracy of the new method increased by 25%,indicating that the new method can make the discriminant results more rapid and accurate.