谱聚类-Adaboost 集成数据挖掘算法在岩性识别中的应用

摘要

针对原有岩性分类方法精度较低、泛化能力不足、结果较不稳定以及不符合地质情况的事实，提出基于谱聚类-Adaboost集成算法的数据挖掘技术，应用谱聚类算法对噪音数据不敏感及可收敛到全局最优解的特点，解决样本数据过滤的问题，有效去除数据冗余；依据数据挖掘集成思想中的 Adaboost 集成算法对基分类器 C4.5进行集成优化，将弱分类器提升为强分类器，提升分类能力。通过对某地区498块致密砂岩岩样资料进行处理，结果表明：谱聚类方法的样本筛选能力较交会图方法与经典聚类方法更强；而 Adaboost 集成算法不仅精度较 BP 神经网络等经典分类算法高，而且具有着较强的泛化能力，较好地解决了基分类器存在的稳定性弱、泛化能力差等问题；利用谱聚类去除样本冗余-Adaboost 集成算法判别的思想使得算法的稳定性更高，岩性判别率稳定到81.96%，明显高于其他判别方法；该方法思路新颖，效果较好，可以进行推广。%In view of the fact that the original lithology classification method has low accuracy,insufficient generalization ability, unstable result and is not consistent with the geological conditions,data mining technology based on spectral clustering-Adaboost integration algorithm was proposed.Through the data processing about 498 sandstone samples,the results show that the sample screening ability of spectral clustering method is better than that of the classical clustering method and intersection graph method. Adaboost integrated algorithm has a strong ability of generalization.It solves the problem of the base classifiers.The stability of the algorithm is improved by using the spectral clustering algorithm to remove the redundancies of samples and using the Ada-boost integrated algorithm to discriminate.Lithology discrimination rate is stable to 81.96% and significantly higher than other discriminant method.The above shows that the method is novel and its effect is good,which shows that it can be popularized.