Electrical and Computer Complex, Maleke Ashtar University of Technology, Tehran, Iran.
Environ Monit Assess. 2018 Jul 12;190(8):464. doi: 10.1007/s10661-018-6837-6.
Seismic events such as earthquakes are one of the most important issues in the field of geology. Meanwhile, less attention has been paid to micro-seismic events, despite the high number of earthquakes. Earthquakes, regardless of their size, affect human life; therefore, their detection and management is considered an important issue. For this purpose, experts developed seismic arrays as a system of linked seismometers. These systems equipped with sensors and seismographs are able to receive a range of waves from the earth, which are then sent to the central seismic station for analysis. So far, many tools and methods have been devised to analyze seismic data. However, the dominant method in most seismic mechanisms is trigger function, based on STA/LTA (short-time-average through long-time-average trigger). These mechanisms have considerable threshold in terms of earthquake range, so many micro-events are ignored as noise. Generally, in this field of geology, computer science techniques have been used to detect and classify these events. Statistical methods such as kurtosis, variance, and skewness can be applied to understand the changes in the signal curves of geophones in a seismic event, thereby helping in the initial detection of fuzzy features. According to the last 3 years' reports of global data mining agencies such as Rexer, KDnugget, and Gartner, Rapid Miner is one of the most popular tools for data mining in recent years. Furthermore, these institutions considered artificial neural networks, especially multilayer perceptron (MLP) and base radial function (RBF), to be among the most successful algorithms for detection and classification of stream data. In this research, the recorded data from several seismic experiments has been classified by a hybrid model. Hence, the present study was aimed to enhance the authenticity of data based on the application of effective variables. This was undertaken through use of a fuzzy method and an integrated neural network algorithm, involving MLP perceptron and radial network of RBF in the form of a collective learning system, in order to identify seismic events on a small scale. Based on the results, in comparison to basic methods, the proposed method significantly improved using the actual error and root-mean-square error (RMSE) criteria.
地震等地震事件是地质学领域最重要的问题之一。同时，尽管地震数量众多，但人们对微地震事件的关注较少。地震无论大小，都会影响人类生活；因此，检测和管理地震被认为是一个重要问题。为此，专家们开发了地震阵列作为一个连接地震仪的系统。这些系统配备了传感器和地震仪，能够接收来自地球的一系列波，然后将这些波发送到中央地震站进行分析。到目前为止，已经设计了许多工具和方法来分析地震数据。然而，在大多数地震机制中，主导方法是基于 STA/LTA（短时间平均到长时间平均触发）的触发功能。这些机制在地震范围方面有相当大的阈值，因此许多微震事件被忽略为噪声。通常，在地质学领域，计算机科学技术已被用于检测和分类这些事件。可以应用峭度、方差和偏度等统计方法来了解地震事件中地震检波器信号曲线的变化，从而有助于初始检测模糊特征。根据 Rexer、KDnugget 和 Gartner 等全球数据挖掘机构过去 3 年的报告，RapidMiner 是近年来最受欢迎的数据挖掘工具之一。此外，这些机构认为人工神经网络，特别是多层感知器 (MLP) 和基本径向函数 (RBF)，是用于检测和分类流数据的最成功算法之一。在这项研究中，通过混合模型对来自几次地震实验的记录数据进行了分类。因此，本研究旨在通过应用有效变量来增强数据的真实性。这是通过使用模糊方法和集成神经网络算法来实现的，该算法涉及 MLP 感知器和 RBF 的径向网络，以集体学习系统的形式，旨在识别小规模的地震事件。基于结果，与基本方法相比，所提出的方法在使用实际误差和均方根误差 (RMSE) 标准时显著提高。
