针对冲击地压预警困难这一难题,基于地音监测提出一种新的前兆信息辨识模型及方法。在固定大小的时间窗口内对地音监测信号进行时频域特征提取,得到11个表征冲击地压灾害前兆的多维特征向量,以实际地音监测数据为训练样本,基于SVM理论建立冲击地压多参量前兆信息辨识模型;提出一种新的SVM学习方法,用于解决工程实际应用中的大规模不平衡数据集训练问题,提高SVM分类准确率及速度。利用地音实测数据作为学习样本对支持向量机进行训练,建立相应的前兆辨识模型进行辨识,准确率达到93.87%。实验分析表明,这种方法有效可靠,样本辨识速度快,能够满足在线监测要求,具有工程应用前景。 Aiming at the difficult problem of early warning of rock burst, a new model and method of precursor information identification are proposed based on geotextile monitoring. Time-frequency feature extraction of geotextile signals in a fixed-size time window to obtain 11 multidimensional feature vectors that characterize the precursory impact of rock burst. Using the actual sound monitoring data as a training sample, based on the SVM theory, A new SVM learning method is proposed to solve the large-scale imbalanced data set training problem in engineering practice and to improve the classification accuracy and speed of SVM. Using the measured data of geotextile as a learning sample to train SVM, the corresponding precursor identification model was established for identification, with an accuracy rate of 93.87%. Experimental analysis shows that this method is effective and reliable, the sample identification speed is fast, it can meet the requirements of on-line monitoring and has the prospect of engineering application.