基于CNN和BiLSTM的地震检测漏报优化算法

Optimization algorithm for seismic detection underreporting based on CNN and BiLSTM

随着定位技术的不断发展,高频GPS技术逐渐成为地震研究领域的热点,可为地震预警工作做出一定补充。本文针对地震检出过程中的漏报问题引入深度学习技术,提出融合区域特征的CNN-BiLSTM分类模型,对新西兰地区高频GPS时间序列进行分析。首先探究小波基及分解层数对降噪效果的影响;其次对高频GPS数据进行归一化、降噪等预处理,并训练融合区域特征的分类模型,实现地震检测。通过与预测模型及单独的CNN、BiLSTM模型进行对比,表明本文模型可有效降低漏检率,具有一定应用价值。

With the continuous development of positioning technology,high-frequency GPS technology has gradually become a hot spot in the field of earthquake research.At the same time,it can make a certain supplement for earthquake early warning.Aiming at the problem of report miss in the process of earthquake detection,the deep learning technology is introduced,and a CNN Bi-LSTM classification model integrating regional characteristics is proposed to analyze the high-frequency GPS time series in New Zealand.Firstly,the influence of wavelet base and decomposition layers on the noise reduction effect is explored.Secondly,the high-frequency GPS data is preprocessed such as normalization and noise reduction,and the classification model integrating regional characteristics is trained to realize seismic detection.Finally,by comparing with different prediction models,the results show that the model in this paper can effectively reduce the report miss rate and has promising application value.