论文部分内容阅读
尽管概率地震危险性分析的方法框架已日趋完善,但是对所关心的场地地面运动预测模型的选择仍是一个主要的问题。信息论提供了强有力的理论框架,可以很好地引导此选择过程。从信息理论视角来看,可以用模型相关的信息丢失(Kullback-Leibler距离)和有物理意义的单位(bit)表达模型的适合程度。与假设检验不同,信息理论模型的选择不需要特殊决定有关的显著性水平,也不需要模型互相排斥和完全穷举。可以通过模型观测结果的对数似然值的统计期望值估算关键要素Kullback-Leibler距离值。在本研究中,对一组反应谱和宏观地震烈度的模拟观测,说明了基于Kullback-Leibler距离差异的数据驱动地面运动模型的选择。通过信息论可以统一处理两个量。使用Abrahamson和Silva(1997)的地面运动模型生成的数据集将基于模型选择的Kull-back-Leibler距离应用于实际数据,说明了数据驱动模型选择信息理论角度较之先前的尝试(例如Scherbaumetal,2004)的优越性。
Although the methodological framework for probabilistic seismic hazard analysis has matured, the selection of ground motion prediction models for the sites of interest remains a major issue. Information theory provides a powerful theoretical framework that can guide this selection process well. From the perspective of information theory, model suitability can be expressed by model-dependent loss of information (Kullback-Leibler distance) and physically significant bits (bits). Unlike hypothesis testing, the choice of model for the theory of information does not require a special level of significance to be decided, nor does the model be mutually exclusive or exhaustive. The Kullback-Leibler distance value of the key element can be estimated from the statistical expectation of the log-likelihood values of the model observations. In this study, the simulation of a set of response spectra and macro-seismic intensity shows the choice of data-driven ground motion model based on Kullback-Leibler distance differences. Through information theory can handle a unified two. Applying Kull-back-Leibler distance based on model selection to actual data using datasets generated by Abrahamson and Silva (1997) ground motion models illustrates the data-driven model selection information theory perspective compared to previous attempts (eg, Scherbaumetal, 2004 ) The superiority.