神经网络在短期地震跟踪预报中的应用

APPLICATION OF NEURAL NETWORKS IN EARTHQUAKE PURSUIT PREDICTION

本文研究了神经网络方法在基于地震活动性指标的中短期（半年至一年）地震预报和基于非测震学前兆异常从属函数的短期（三个月之内）地震预报中的应用。选用含一个或两个中间层的前向神经网络模型，并采用与之相适应的ＢＰ算法。以华北地区多年的地震活动性资料和首都圈及其邻近地区的短水准、地电阻率、地磁总强度、水位、水氡含量等前兆观测手段的８０余个台项的多年观测资料为基础，对神经网络方法在以上两方面的应用作出了实际计算、分析与检验。对一些大地震的发生时段及震级的预测检验效果较好。本文所得结果表明，神经网络在短期地震跟踪预报中有一定应用价值，值得进一步探讨与推广。

The application of neural networks method in mid-short-term (half to one year) earthquake prediction based on the indices of seismic activity and in short-term (three months) earthquake prediction based on the membership functions of nonseismic precursory anomalies have been studied. The forward -type model of neural networks (NN) containing one or two medium layers was taken, and the BP algorithm was applied correspondingly. On the basis of long-period observational data of seismicity in North China region and short-range levelling, geoelectrical resistivity, geomagnetic total intensity,water level, radon content in groundwater etc. from more than 80 stations/kinds in the Capital Circleand its vicinity, the actual calculations, analyses and checks for applications of NN in both aspects mentioned above were made. The check results of occurrence times and magnitudes of several large earthquakes were satisfactory. Therefore, the results obtained show that the neural networks method/tech-nology can be applied to the earthquake pursuit prediction more-effectivily and prospectivily.