基于BP-Adaboost方法的天然地震和人工爆炸事件波形信号分类识别研究

发布时间：2018-03-23 08:37

本文选题：分类识别　切入点：地震波形信号　出处：《地震工程学报》2017年03期 　论文类型：期刊论文

【摘要】：BP神经网络和支持向量机(SVM)是两种主流的分类识别方法,用于天然地震和人工爆炸事件波形信号分类识别时取得了较好的效果。但BP神经网络存在易陷入局部最优及隐层数和隐层节点数与训练样本数据密切相关而无法有效预先确定;而支持向量机(SVM)方法则缺乏有效手段来选取合适的核函数,从中不能很好地扩展到多分类。针对天然地震和人工爆炸事件波形信号的分类识别问题,文中将上述两种方法和集成学习——BP-Adaboost方法进行了对比实验研究。据对所选用的地震、爆炸事件波形信号数据集的分类识别结果表明,BP-Adaboost方法得到了98%以上的正确识别率,并且具有较好的泛化能力。相较于BP神经网络和PCA-SVM方法,BP-Adaboost方法对于数据集的划分和识别结果具有更好的鲁棒性,应用于天然地震和人工爆炸事件波形信号分类识别时,可取得更好的识别效果。同时,结合Adaboost方法的原理,阐述了BP-Adaboost方法拥有更好分类结果和泛化能力的原因。
[Abstract]:BP neural network and support vector machine (SVM) are two main classification and recognition methods. The BP neural network is easy to fall into the local optimum and the number of hidden layers and hidden layer nodes are closely related to the training sample data. But the support vector machine (SVM) method lacks the effective means to select the appropriate kernel function, which can not be extended to the multi-classification. In view of the problem of classification and recognition of waveform signals of natural earthquake and artificial explosion events, the support vector machine (SVM) method can not be extended to multiple classification. In this paper, the two methods mentioned above are compared with the integrated learning-BP-Adaboost method. According to the classification and recognition results of the selected seismic and explosion waveform data sets, the BP-Adaboost method has obtained a correct recognition rate of more than 98%. Compared with BP neural network and PCA-SVM method, BP-Adaboost method is more robust to data set partition and recognition results. At the same time, combined with the principle of Adaboost method, the reason why BP-Adaboost method has better classification result and generalization ability is expounded.
【作者单位】：广西师范大学计算机科学与信息工程学院;
【基金】：国家自然科学基金(41264001)
【分类号】：P315;TP18

【参考文献】