基于卷积神经网络的冲击地压预警方法研究*

为探究深度学习在冲击地压预警方面的应用前景，以新疆某冲击地压矿井为研究背景，将深度学习和专家评判运用到微震数据分析中，基于卷积神经网络构建冲击地压预警模型。充分利用一维卷积神经网络对时序数据有较强特征提取能力的优势，以微震数据及其特征参数作为输入，以专家评判值作为标签，借助Python-Keras框架实现冲击地压预警模型的构建和训练。研究结果表明:模型预警效果并不随着训练迭代次数的增加而逐渐最优，存在最优迭代次数，对于所建模型当迭代次数为30时测试集的冲击危险预测结果与专家评判结果基本吻合，同时说明模型可以较好地学习专家评判经验实现冲击地压预警。研究表明所建模型对研究时段内发生的5次大能量矿震事件均进行预警，其准确度较高，具有现场实际应用价值。

Research on early-warning method of rockburst based on convolutional neural network

In order to explore the application prospect of deep learning in the early-warning of rockburst,taking a rockburst mine in Xinjiang as the research background,the deep learning and expert evaluation were applied in the microseismic (MS) data analysis,and an early-warning model of rockburst was constructed based on the convolutional neural network.Taking advantage of the strong feature extraction ability of one-dimensional convolutional neural network (1DCNN) for time series data,using MS data and its characteristic parameters as the input and the expert evaluation values as the label,the construction and training of rockburst early-warning model were realized with the Python-Keras framework.The results showed that the early-warning effect of the model was not gradually optimal with the increase of the number of training iteration times,and there existed the optimal number of iteration times.For the established model,when the number of iteration times was 30,the prediction results of rockburst risk by test set were basically consistent with the expert evaluation results,which also indicated that the model could learn from the expert evaluation experience to realize the early-warning of rockburst.It showed that the established model gave early-warning to all the 5 high-energy mine earthquake events in the study period,and the early-warning accuracy was high,which has the value of field application.