基于遗传算法地震短期综合预报分类系统在天山地震带的应用

THE APPLICATION OF THE CLASSIFYING SYSTEM OF SHORT -TERM COMPREHENSIVE EARTHQUAKE PREDICTION BASED ON THE GENETIC ALGORITHMS IN TIANSHAN MOUNTAIN SEISMIC ZONE

应用"基于遗传算法的地震预报分类系统"建立天山地震带8个区的地震短期综合预报模型。在前人研究基础上对此分类系统的应用方法做了以下改进:以半年或一年样本窗内的地震活动资料预报未来3个月最大地震;采用阈值划分异常,区分高值和低值异常两类参数,再用异常持续时间进行地震环境信息编码;在研究区内,根据一年和半年样本窗和震级错半级的分档形式,形成4种地震环境信息编码方式。计算机自动运行此系统程序,寻找各自的预报模型。对预留样本的检验按震级区间进行规则对应率、虚报率和地震对应率、漏报率的效能评价,结果表明南天山各区检验情况好于北天山;对北天山预测出5级以上、南天山6级以上地震以及能给出有效重叠震级区间的模型加权后认为南天山中西段、柯坪块体区检验情况较好。结果还表明半年样本窗检验效能普遍高于一年窗,对较低震级的检验情况一般要好于较高震级。

The classifying system of earthquake prediction based on the genetic algorithms is applied to set up the models of the short-term comprehensive prediction of 8 areas in Tianshan Mountain seismic zone. We improved the method as followed based on primal studies. The largest earthquake is predicted in three months by applying the seismic activity data of the sample windows in six months or one year. We applied a threshold to mark the anomaly, and distinguish two types of anomalous parameters (i.e. the high value and the low value). Then translated the codes of seismic circumstance information using lasting time of abnormity. In study section, four codes are given according to the sample windows of one year and half a year and the grading form of magnitudes with a half step. And the computer will automatically run the system program. For inspection of reserved samples, we evaluate efficiency of rule corresponding rate, virtual prediction rate and seismic corresponding rate, missing prediction rate according the magnitude. The study result shows that the inspection efficiency of Southern Tianshan Mountain is better than that of Northern Tianshan Mountain. And we think that the inspection efficiencies of the mid-west section of Southern Tianshan Mountain and Keping block are very well for Northern Tianshan Mountain predicted earthquakes with more than M_S5, Southern Tianshan Mountain more than M_S6 and after weighting the model able to giving efficiency overlapped magnitude section. The result as well shows that the inspection efficiency of a half-year sample window is better than one of a year window. And the inspection of magnitude with lower is better than one of one with higher.