Percolation theory as a tool for imitation of fracture process in rocks

The statistical models of earthquake focal processes covering topological, kinetic, energetic and scaling aspects of the phenomenon are considered on the basis of the percolation and reliability theories. It is shown that these models do not contradict the basic experimental data of physics of fracture and seismology. Some predictive signs are also considered.     

Stratigraphic evidence for millennial-scale temporal clustering of earthquakes on a continental-interior fault: Holocene Mississippi River floodplain deposits, New Madrid seismic zone, USA

The earthquake cycles that characterize continental-interior areas that are far from active plate boundaries have proven highly cryptic and difficult to resolve. We used a novel paleoseismic proxy to address this issue. Namely, we reconstructed Holocene Mississippi River channels from maps of floodplain strata in order to identify channel perturbations reflective of major displacement events on the high-hazard and mid-plate Reelfoot thrust fault, New Madrid seismic zone, U.S.A. Only three discrete slip events are currently documented for the Reelfoot fault ( AD 900,  AD 1450, and AD 1812). This study extends this record and, thus, illustrates the utility of stratigraphic proxies as paleoseismic tools. We concurrently offer here some of the first quantified response times for tectonically induced channel pattern changes in large alluvial rivers.

We identified at least two cycles of pervasive meandering that were interrupted by channel-straightening responses occurring upstream of the Reelfoot fault scarp. These straightening responses initiated at 2244 BC +/− 269 to 1620 BC +/− 220 and  AD 900, respectively, and each records initiation of a period of Reelfoot fault slip after millennia of relative tectonic quiescence. The second (or New Madrid) straightening response was triggered by the previously known  AD 900 fault slip event, and this initial low sinuosity has been protracted until the modern day by the latter  AD 1450 and AD 1812 events. The first (or Bondurant) straightening response began a period of several hundred to  1400 years of low river sinuosity which evidences a similar period of multiple recurrent displacement events on the Reelfoot fault. These Bondurant events predate the existing paleoseismic record for the Reelfoot fault.

These data offer initial evidence that slip events on the Reelfoot fault were temporally clustered on millennial scales and, thus, offers the first direct evidence for millennial-scale clustering of earthquakes on a continental-interior fault. This carries additional ramifications. Namely, faults that have been quiescent and non-hazardous for millennia could re-enter an enduring period of recurrent hazardous earthquakes with little warning. Likewise, the Reelfoot fault also reveals evidence of temporal clustering of earthquakes on short-term cycles (months), as well as evidence for longer-term reactivation cycles (10⁴–10⁶ years). This introduces the possibility that temporal clustering could be hierarchical on some continental-interior faults.     

混沌退火Hopfield网络及其在储层聚类分析中的应用

将一种基于混沌退火的Hopfield神经网络(ACHN)应用到储层聚类分析中。这里主要阐述了ACHN聚类的基本原理、混沌神经元的标定、ACHN聚类的策略、先验信息的约束以及混沌退火搜索。实际砂砾岩储层聚类分析表明:这种网络能克服普通Hopfield网络陷入局部最优的不足,得到更准确、精细的聚类结果。     

模糊聚类实现岩性自动划分

利用测井数据中的自然伽玛、中子、声波和密度测井曲线所蕴含的岩性信息,用模糊聚类算法实现岩性的自动划分。该方法识别准确度高,可靠性强。通过对川西坳陷十八口井的实际应用,此方法在低孔低渗的致密碎屑岩岩性划分中取得了良好的效果。     

Analysis of Shallow and Deep Earthquake Doublets in the Fiji-Tonga-Kermadec Region

For the Fiji-Tonga-Kermadec area and for the period from January 1977 to July 2003, the Harvard CMT catalogue lists 1022 shallow, 410 intermediate and 633 deep earthquakes of moment magnitude from 4.9 to 8.0. The magnitude threshold, above which the catalogue is complete, is 5.3–5.4, and the number of earthquakes of magnitude above this value is 691 for shallow, 329 for intermediate and 476 for deep events, respectively. The proportion of earthquakes associated with doublets and multiplets against the total number of earthquakes is approximately the same in both data sets and therefore all earthquake pairs were considered regardless of their magnitude. We investigated all the pairs of earthquakes that occurred at a centroid distance of less than 40, 60 or 90 km from each other and within a time interval of 200, 300 or 450 days, depending on their magnitude. We found 208 pairs of shallow, 31 of intermediate and 92 of deep events. To ascertain whether these earthquakes in pairs are not connected by chance, the possibility of their occurrence in an uncorrelated Poissonian catalogue was considered. It was assumed that in such a catalogue the inter-event time is exponentially distributed, the earthquake magnitude follows the Gutenberg-Richter relation, and the distribution of centroid distances between the events in pairs is controlled by its non-parametric kernel estimate. The probability of the appearance of the observed proportion of doublets of shallow earthquakes in the Poissonian catalogue was found to be very low. The low probability of occurrence in a semi-random catalogue, created by randomising centroid locations in the actual data set, also indicates major importance of the distance criterion used for a doublet specification. In general, shallow earthquakes tend to form pairs at shorter distances and within shorter time intervals than deep earthquakes. Both the distance and the time intervals do not depend on the magnitude of involved events. The largest number of pairs of deep earthquakes is observed at a depth of about 600 km, and the proportion of deep events associated with doublets against the number of all events increases with depth. From comparison of the focal mechanism of earthquakes in pairs, measured by the 3-D rotation angle, it follows that deep earthquakes forming pairs have a more diverse focal mechanism than shallow events; the rotation angle for three quarters of shallow pairs and only for about one third of deep pairs is reasonably small. The azimuth between two events forming a doublet is in about 60–65% of cases close to the strike of one of nodal planes of the first or the second event.     

Simultaneous determination of aquifer parameters and zone structures with fuzzy c-means clustering and meta-heuristic harmony search algorithm

This study proposes an inverse solution algorithm through which both the aquifer parameters and the zone structure of these parameters can be determined based on a given set of observations on piezometric heads. In the zone structure identification problem fuzzy c-means (FCM) clustering method is used. The association of the zone structure with the transmissivity distribution is accomplished through an optimization model. The meta-heuristic harmony search (HS) algorithm, which is conceptualized using the musical process of searching for a perfect state of harmony, is used as an optimization technique. The optimum parameter zone structure is identified based on three criteria which are the residual error, parameter uncertainty, and structure discrimination. A numerical example given in the literature is solved to demonstrate the performance of the proposed algorithm. Also, a sensitivity analysis is performed to test the performance of the HS algorithm for different sets of solution parameters. Results indicate that the proposed solution algorithm is an effective way in the simultaneous identification of aquifer parameters and their corresponding zone structures.     

Abell团相关性-富度关系的统计研究

本文对D≤4,R≥0的Abell团样品和整个R≥3的Abell团样品的空间两点相关函数做了统计分析,结果表明R≥0,R=0和R≥3的Abell团的相关函数分别为175r~(-1.8),150r~(-1.8)和1900r~(-1.8)利用已有的R≥1和R≥2团的相关函数的统计结果,我们得到了Abell团的相关性幅度α与富度N之间的关系:α∝N~(1.6)。在相关性幅度的统计误差所允许的范围内,α与N之间的关系可以在宇宙弦理论中得到解释。     

基于自适应膨胀因子的模糊覆盖分级方法

分级界限的确定一直都是学者们关注的焦点,而如何确定分级数则往往被忽略了。根据模糊覆盖和膨胀因子的定义,提出了一种确定1维数据分级数的分级方法:首先通过覆盖半径做一个模糊覆盖,利用自膨胀因子对这个覆盖进行扩张,直到不能再向外扩张为止,这一级就确定了;然后进行下一级的确定。该方法不仅可以自动确定数据的分级数,而且克服了分级结果对参数的敏感性。最后用实验证明了方法的有效性。     

基于MRF随机场和广义混合模型的遥感图像分级聚类

有限混合模型FM的分级聚类已广泛应用于不同领域,然而,它的计算复杂度与观测数据的平方成正比,因此,在海量数据方面的应用就受到了限制。另一方面,多光谱图像数据中同时包含有空间和光谱两类信息,但大多数基于像素的多光谱图像聚类方法,仅使用了其频谱信息而忽视了空间信息。本文提出了一种新的基于广义有限混合模型GFM的分级聚类方法,该算法把MRF随机场和GFM模型结合在一起,分类数可以通过PLIC准则自动确定。算法在执行过程中,采用K均值聚类方式获得过分类图像,分级聚类从过分类图像开始,代替原来从单点类开始的方式,这样可以方便获取GFM模型成分密度的初始参数。最后,采用由Gibbs采样器生成的仿真测试图对算法的精度进行了定量评价,通过与K均值聚类和FM聚类的比较说明了本文算法的优越性,同时用荷兰Flevoland农业地区的极化SAR图像验证了本文算法的有效性。