【摘要】 以准确区分天然地震与非天然地震为目标，构建了一种基于一维卷积和残差结构的神经网络模型：ResNet-1D。该模型利用不同长度卷积核的卷积层、最大池化构成的池化层和残差结构自动提取三分量地震记录特征，采用适应性矩估计方法(Adams)作为优化参数，利用线性判别函数(Linear)实现天然地震与非天然地震区分。以2008—2020年中国地震台网中心统一编目报告的天然地震和非天然地震共40 000条速度记录，随机划分为6∶2∶2的比例构建训练数据集、验证数据集和测试数据集。研究结果表明：天然地震和非天然地震的分类准确率分别为92.65%和94.30%,与传统机器学习方法比较，ResNet-1D模型在准确率、精确率、召回率和F1分数的测试结果均有明显提升，有效地提高了天然地震和非天然地震识别的准确性。同时，震级和震中距的变化对模型分类准确率都有影响，具体表现为震级越高，准确率越低；震中距越大，准确率越低。文中提出的模型具有更高的准确性，可为地震监测中的天然地震与非天然地震准确区分提供技术支撑。更多还原

【Abstract】 Aiming to accurately differentiate between natural and non-natural earthquakes, a neural network model based on one-dimensional convolution and residual structures, named ResNet-1D, was constructed. This model automatically extracts features from three-component seismic records using convolutional layers with convolutional kernels of different lengths, pooling layers composed of max-pooling, and residual structures. The adaptive moment estimation method(Adams) is used to optimize parameters, and a linear discriminant function(Linear) is applied to distinguish between natural and non-natural earthquakes. Using 40 000 velocity records of natural and non-natural earthquakes, compiled by the China Earthquake Networks Center from 2008 to 2020, the data was randomly divided into training, validation, and test datasets in a 6∶2∶2 ratio. The test results show that the classification accuracy for natural and non-natural earthquakes is 92.65% and 94.30%, respectively. Compared with traditional machine learning methods, the ResNet-1D model significantly improves the test results in terms of accuracy, precision, recall, and F1 score, effectively enhancing the accuracy of identifying natural and non-natural earthquakes. Moreover, variations in magnitude and epicentral distance also affect the classification accuracy of the model, with higher magnitudes and greater distances resulting in lower accuracy. The model proposed in this paper offers higher accuracy and provides technical support for accurately distinguishing between natural and non-natural earthquakes in seismic monitoring.更多还原