基于快速支持向量机算法的灌浆地层识别

Grouting stratum identification with fast support vector machine

针对灌浆地层裂隙存在不确定性变化的特点,提出基于支持向量机(SVM)的灌浆地层智能识别方法,以提高地层识别能力。为了提高支持向量机模型的运算速度,采用几何方法求取支持向量,避免了二次规划算法求解。该算法根据支持向量的几何分布特点,从距离最近的样本点开始,通过不断地寻找违反KKT条件的样本点来找出支持向量。最后,选取大理岩石(较完整地层)和砂质板岩地层的部分灌浆实验数据样本建立灌浆地层的分类模型,用其他未训练的同分布的新数据进行地层识别验证。仿真结果表明,改进支持向量机分类方法简单有效,与神经网络方法相比有更强的泛化能力和更快的运算速度。

A support vector machine （SVM） approach was proposed for uncertain grouting stratum to increase the grouting stratum identification capability. In order to improve SVM algorithm＇s iterative efficiency, a geometric algorithm of support vector was proposed to avoid resolving quadratic programming algorithm. The principle of the new algorithm is as follows： starting from two closest points of the opposite classes to seek the support vectors accumulatively, and seeking the vectors which are the violators of KKT condition as support vectors. Finally, selecting group classification data of fracture rock and gritstone stratum in the real grouting project, a part of them was used to train set of SVM, and others were used to check the classification effect. Simulation results show that SVM method is simple and effective, and it has strong robust ability and faster computing velocity compared with neural network method.