摘要
将投影寻踪(PP)与高维时间序列分析结合起来,建立了地震PP综合预测模型.并选取祁连山地区作研究区,做了未来三个月内最大震级的短期预测.经展望式检验,合格率(预测误差≤0.5或≤10%)≥80%.文中根据实际需要提出建立一定震级门限(M≥4.0)的预测具有更好的效果.本预测模型还可增加自变量、维数和改变时间尺度,并运用到其他领域中.
Combining Projection Pursuit(PP) and highdimensional time series analysis,the synthetic earthquake prediction model of highdimensional PP time series is built.Selecting Qilian mountain region as research object,its shortterm prediction model of the maximum magnitude within three months is made.The qualified rate of prediction test is more than 80% (Prediction error≤0.5 or ≤10%).In this paper,it is considered that the model with certain magnitude threshold (MS≥4.0) is better accordding to the real cases.It is available for the model to increase parameters,dimensions,change prediction periods,and generalize to prediction in other fields.
引文
[1]成平,李国英.投影寻踪———一种新兴的统计方法[J].应用概率统计,1986,(3):138—143.
[2]李祚泳.投影寻踪的理论及应用进展[J].大自然探索,1998,(1):47—50.
[3]朱令人,周仕勇,邓传玲.地震综合预报的新方法———投影寻踪回归[J].地震学报,1994,16(增刊):1—9.
[4]周仕勇,朱令人,邓传玲.PP聚类在震群分析中的应用研究[J].地震学报,1995,17(3):312—321.
[5]赵翠萍,周仕勇.新疆主要地震区PP回归综合预报模型研究[J].西北地震学报,1999,21(1):37—43.
[6]邓传玲.SMART多重平滑回归技术的原理及计算软件[J].八一农学院学报,1988,(4):47—55.
[7]吴佳翼,曹学锋.地震活动性的定量化问题[J].地震,1983,(6):13—16.
[8]周仕勇,王海涛.地震空间分布的带状集中度Cb[J].地震,1991,(2):59—64.
[9] 谷继成,魏富胜.论地震活动性的定量化:地震活动度[J].中国地震,1987,10(增刊):12—22.
[10] 朱令人,周仕勇,龚宇清,等.信息量在地震活动研究中的应用[J].西北地震学报,1991,13(3):15—20.
[11] 朱令人,龚宇清,等.玛纳斯地震活动信息熵及分维的动态变化[J].内陆地震,1990,4(3):193—200.
[12] 朱令人,王海涛,杨马陵.乌鲁木齐和阿拉木图地区部分地震前固体潮小震调制比异常[J].内陆地震1991,5(1):11—16.
[13]ATWalden.Spatialclustering:usingSimpleSummariesofSeismicDatatoFindtheEdgefoanOil field[J].AppliedStatistics,1994,43:335—345.
[14] JHFriedman,WStuetzle,ProjectionPursuitRegression[J].Amer.Statist.Assor.,1981,76:817—823.
[15] SBorra,ADCiaccio.Non parametricRegressionModelsfortheConjointAnalysisofQualitativeandQuantitativedata[A].See:DataScienceClassificationandrelatedmethodsVIConferenceofIFCS[C].Rome,1998.21—24.