论文部分内容阅读
提出了用反向传播智能网作为检测器来识别宽频带震相的一种方法。通过对3种反向传播智能检测器(长周期、中周期和短周期)的结果进行综合判断,认为这种方法既有短周期检测器准确性高的特点又有长周期检测器误警率低的特点。我们举例证明了适当地对数据进行预处理和后处理有助于改进系统的性能。本文也对反向传播智能网的结构和参数的设定进行了讨论。我们用研制成功的反向传播智能网检测器对美国地震学联合研究协会地震台网的1254张宽频带地震图上的地震事件进行了初至检测,希望能用这些(?)时资料进行地幔结构的层析成像研究。结果表明:1254张地震图中95%以上可识别出初至。自动识别的走时精确度尚可,85%以上的走时误差小于1 s,约80%的延时误差小于0.5s。
A method of identifying broadband phases by using back propagation intelligent network as a detector is proposed. By comprehensively judging the results of three types of backpropagation intelligent detectors (long period, medium period and short period), it is considered that this method not only has the characteristics of high accuracy of short period detector but also false detection rate of long period detector Low characteristics. We show by example that proper preprocessing and postprocessing of data can help to improve system performance. This article also discusses the structure and parameters of backpropagation intelligent network. We conducted a first arrival detection of the seismic events on the 1254 wideband seismograms of the Seismological Research Association Seismological Network using a developed backpropagation intelligent network detector, hoping to use these data to conduct mantle Structural tomography research. The results show that more than 95% of the 1254 seismograms can recognize the first arrival. The accuracy of travel time of automatic recognition is acceptable, the travel time error of more than 85% is less than 1 s, and the delay error of about 80% is less than 0.5s.