【摘要】 本文提出了一种基于卷积降噪自编码器和Softmax回归的微地震定位方法.该方法首先将微地震数据输入到卷积降噪自编码器中进行随机噪声压制，利用卷积降噪自编码器的编码器提取微地震数据的鲁棒性特征.然后根据震源的地理位置，对每个微地震数据生成多个独立的位置标签.使用带有震源位置标签的微地震特征训练多输出的Softmax分类器模型，同时预测一个输入微地震数据的多个位置标签，进而获得精确的震源位置.合成地震记录的实验结果表明，该方法能够准确快速地定位低信噪比的微地震事件.更多还原

【Abstract】 We present a microseismic location method based on convolutional denoising auto-encoder and softmax regression. Firstly, the microseismic data are input to the convolutional denoising auto-encoder for random noise suppression, and the encoder of the convolutional denoising auto-encoder is used to extract the robustness characteristics of the microseismic data. Then, multiple independent location labels are generated for each microseismic data based on the geographic locations of the sources. Finally, a Softmax classifier model with multiple outputs is trained using microseismic features with location labels. The model can simultaneously predict multiple location labels for the given microseismic data and thus determine the precise source location. The experimental results of synthetic microseismic data show that the proposed method can locate microseismic events with low signal-to-noise ratio accurately and quickly.更多还原