基于ACO优化SVM在测井岩性识别中的研究

发布时间：2018-04-21 01:25

本文选题：支持向量机 + 测井岩性识别　；参考：《东北石油大学》2015年硕士论文

【摘要】：在油气勘探的过程中,随着深度的不断加深,在地形较复杂的地方,使用传统的方法很难快速准确的分辨出油气位置,但处理此问题可使用岩性识别方法。岩性识别过程是一个非常典型的、具有高维特性的、很难解决的非线性的模式识别过程,而支持向量机(Support Vector Machine,简称SVM)正是解决此问题的最优方法之一,其能够很好地处理小样本、灵活的转换非线性、强悍的应对高维模式识别等问题。本文首先介绍了蚁群算法的理论和支持向量机的理论;其次,由于基于支持向量机参数的选取直接影响到支持向量机分类性能的好坏,因此,介绍了几种支持向量机参数寻优的方法,如网格搜索法、双线性搜索法、穷举法、遗传算法、粒子群算法等,重点介绍了将交叉验证法与蚁群算法的有效结合的新蚁群算法,将新蚁群算法优化支持向量机同蚁群算法优化支持向量机进行了比较,得知前者不但缩短了优化SVM的时间,也提高了分类准确率;最后,针对大规模数据分类问题,传统的支持向量机表现出很多的不足,因此,本文对传统SVM进行了改进-最近邻支持向量机,使用岩性测井数据训练改进后的支持向量机和传统的支持向量机,将实验结果进行比较,其结果表明:针对大规模数据分类,最近邻支持向量机表现出一定的优势。
[Abstract]:In the process of oil and gas exploration, along with the deepening of depth, it is difficult to distinguish the oil and gas position quickly and accurately by using the traditional method in the places where the topography is more complicated, but the lithologic identification method can be used to deal with this problem. The lithologic recognition process is a very typical, high dimensional and difficult nonlinear pattern recognition process. Support Vector Machine (SVM) is one of the best methods to solve this problem. It can deal with small samples, convert nonlinearity flexibly, and deal with high dimensional pattern recognition. This paper first introduces the theory of ant colony algorithm and support vector machine. Secondly, because the selection of parameters based on support vector machine directly affects the classification performance of support vector machine, This paper introduces several methods for parameter optimization of support vector machines, such as grid search, bilinear search, exhaustive, genetic algorithm, particle swarm optimization, etc. Comparing the new ant colony optimization support vector machine with the ant colony optimization support vector machine, we know that the former not only shortens the time of optimizing SVM, but also improves the classification accuracy. Finally, aiming at the large-scale data classification problem, The traditional support vector machine (SVM) shows a lot of shortcomings. Therefore, this paper improves the traditional SVM (nearest neighbor support vector machine), uses lithologic logging data to train the improved support vector machine (SVM) and the traditional support vector machine (SVM). The experimental results show that the nearest neighbor support vector machine has some advantages for large-scale data classification.
【学位授予单位】：东北石油大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：P631.81;TP18

【参考文献】