基于机器学习算法的脉冲型地震动识别
发布时间:2018-07-08 16:08
本文选题:脉冲型地震动 + S变换 ; 参考:《哈尔滨工业大学》2017年硕士论文
【摘要】:已发生的地震灾害统计和研究显示,在相同的震级和场地因素下,相较于普通地震,脉冲型地震对建筑结构造成的毁坏更加严重。这类地震动通常由断层破裂时的前场方向性效应产生。对这些地震动的类型进行准确的识别是地震工程领域的一项基本任务。这有助于创建可靠的脉冲和非脉冲型地震记录数据库,供研究者进行地震危险性概率分析等。本文从人工提取特征和特征学习的角度出发,分别采用支持向量机和深度神经网络建立识别模型。从人工提取特征的角度出发,本文首先采用S变换重构原始地震速度记录中的脉冲时程,根据重构的脉冲时程提取用于后续判别的特征。之后,采用主成分分析去除部分特征的相关性,降低特征的冗余度。然后,依据主成分分析的结果,采用支持向量机判别某地震动是否为脉冲型。结果表明,S变换能够有效的重构原始地震速度记录中的脉冲时程;该方法能够一定程度上有效的判别地震动的类型。采用支持向量机识别地震动的类型,识别过程中特征的提取依赖研究者的经验和专业知识。深度神经网络能够自动地提取特征集用于判别地震动的类型。因此,从特征学习的角度出发,本文首先通过将原始地震动速度时程转为图像,以图像像素的方式作为神经网络模型的输入。之后,通过栈式自编码算法提取用于判别地震动类型的特征,并结合Softmax分类器建立相应的深度神经网络识别模型;采用交叉验证的方式评估该模型的判别效果和稳定性。结果表明,该深度神经网络具备良好的稳定性,能够有效的提取特征集识别地震动的类型。
[Abstract]:According to the statistics and research of earthquake disaster, the damage to building structure caused by pulse earthquake is more serious than that of ordinary earthquake under the same magnitude and site factors. This type of ground motion is usually caused by the front-field directional effect when the fault breaks. Accurate identification of these types of ground motion is a basic task in the field of seismic engineering. This will help to create reliable pulse and non-pulse seismic record databases for researchers to carry out seismic risk probability analysis and so on. In this paper, support vector machine (SVM) and depth neural network (DRNN) are used to establish recognition model from the point of view of feature extraction and feature learning. From the point of view of artificial feature extraction, this paper first uses S transform to reconstruct the pulse time history in the original seismic velocity record, and extracts the feature for subsequent discrimination according to the reconstructed pulse time history. After that, the correlation of some features is removed by principal component analysis (PCA), and the redundancy of feature is reduced. Then, according to the results of principal component analysis, support vector machine (SVM) is used to determine whether a ground motion is of pulse type. The results show that the S-transform can effectively reconstruct the pulse time history in the original seismic velocity record, and the method can effectively distinguish the type of ground motion to a certain extent. Support vector machine (SVM) is used to identify the types of ground motion. The feature extraction in the recognition process depends on the experience and expertise of the researchers. Depth neural networks can automatically extract feature sets to identify the type of ground motion. Therefore, from the point of view of feature learning, this paper firstly transforms the original velocity of ground motion into an image and uses the pixel of the image as the input of the neural network model. After that, the self-coding algorithm of stack is used to extract the features used to distinguish the type of ground motion, and the corresponding recognition model of depth neural network is established by using Softmax classifier, and the discriminant effect and stability of the model are evaluated by cross-verification. The results show that the depth neural network has good stability and can effectively extract feature sets to identify the type of ground motion.
【学位授予单位】:哈尔滨工业大学
【学位级别】:硕士
【学位授予年份】:2017
【分类号】:P315.9
【参考文献】
相关期刊论文 前1条
1 S.R. Hoseini Vaez;M.K. Sharbatdar;G. Ghodrati Amiri;H. Naderpour;A. Kheyroddin;;Dominant pulse simulation of near fault ground motions[J];Earthquake Engineering and Engineering Vibration;2013年02期
,本文编号:2107974
本文链接:https://www.wllwen.com/kejilunwen/diqiudizhi/2107974.html