基于改进SimCNN模型的矿山地震灾害识别研究  

Research on mine earthquake disaster identification based on improved SimCNN model

作　　者：张浩鸣 周煊超 阿那尔[1] Zhang Haoming;Zhou Xuanchao;A Na′er(Earthquake Agency of Inner Mongolia Autonomous Region,Hohhot 010010,China)

机构地区：[1]内蒙古自治区地震局,内蒙古呼和浩特010010

出　　处：《能源与环保》2024年第10期27-33,共7页CHINA ENERGY AND ENVIRONMENTAL PROTECTION

基　　金：内蒙古自治区地震局局长基金课题(2021ZF06);中国地震局地震科技星火计划(XH200501)。

摘　　要：煤矿开采过程中产生的矿山微震会给矿井带来较大的损害。为了对矿山微震进行识别,并提升识别准确率,采用卷积神经网络,并引入了相似特征层来对其进行改进。由于微震数据集的样本量较小,难以满足模型的训练要求,采用迁移学习将地震数据特征迁移到微震数据中,并借此数据集对改进后的卷积神经网络进行预训练。为了进一步提升改进后卷积神经网络的特征获取能力,在现有改进的基础上又引入了长短时记忆网络和注意力机制,以获取数据的时序特征和完整的矿山微震识别模型。研究结果显示,在测试集上,研究设计矿山微震识别模型的耗时平均值为8.216 s,识别准确率最大值和最小值分别为99.87%和93.29%,皆明显优于对比矿山微震识别模型。矿山微震识别模型能够对矿山微震时破碎煤岩的纵波和横波速度进行较为准确的预测。研究设计的矿山微震识别模型性能较好,能够为当下煤矿开采过程中的微震识别提供技术支持。Mine microseisms generated during coal mining can cause significant damage to the mine.In order to identify microseisms in mines and improve accuracy,convolutional neural networks were used in the study,and similar feature layers were introduced to improve it.Due to the small sample size of the microseismic dataset,it is difficult to meet the training requirements of the model.Therefore,the study also adopted transfer learning to transfer the features of seismic data to microseismic data,and used this dataset to pre train the improved convolutional neural network.In order to further enhance the feature acquisition capability of the improved convolutional neural network,research introduced long short-term memory networks and attention mechanisms on the basis of existing improvements to obtain temporal features of data and a complete mining microseismic recognition model.The results showed that on the test set,the average time taken to study and design the mining microseismic identification model was 8.216 seconds,with the maximum and minimum recognition accuracy values of 99.87% and 93.29%,respectively,which were significantly better than the comparison mining microseismic identification model.The mining microseismic identification model can accurately predict the longitudinal and transverse wave velocities of fractured coal and rock during mining microseismic events.The design of a mining microseismic identification model has good performance and can provide technical support for microseismic identification in current coal mining processes.

关 键 词：SimCNN 相似特征层 微震 煤矿 矿山微震识别 

分 类 号：TD76[矿业工程—矿井通风与安全]