基于样式降维聚类的多点地质统计建模算法

评价多点地质统计建模算法的一个重要指标是能否在保证建模质量的同时较好地协调计算效率和内存空间二者的平衡。基于样式的多点地质统计建模算法SIMPAT存在一定不足,使得SIMPAT算法在提出多年后仍然难以实际应用。国外学者提出了SIMPAT算法的改进算法:Filtersim和DisPat,但是在平衡计算效率和内存空间问题上仍存在一定的不足。通过深入剖析SIMPAT算法的原理和特点提出了基于样式降维聚类的多点地质统计建模算法。新算法采用邻近等间距取样法对所有数据样式进行降维聚类处理,把相似的数据样式聚为一类。不同于SIMPAT算法的一次相似度匹配计算,新算法采用2次相似度比较:先比较数据事件与样式类代表性样式的相似度,找到最相似的样式类;再进一步比较数据事件与该样式类中全部数据样式的相似度,进而确定相似度最大的数据样式。以二维和三维实例比较了新算法与传统多点地质统计建模算法SIMPAT、Snesim、Filtersim和DisPat在相同参数条件下的模拟计算效率。结果表明,新算法在保证模拟质量基础上极大提高了基于样式的多点地质统计建模算法的计算效率,并节省内存空间。

Multiple-point geostatistics algorithm based on pattern scale-down cluster

A key evaluation indicator of multiple-point geostatistics modeling algorithm is to ensure modeling quality when harmonizing the equilibrium between computational efficiency and memory space. Simulation of Patterns (SIMPAT) is a pattern-based multiple-point geostatistics algorithm for geological modeling, leading to a low computational efficiency of SIMPAT. Thus, SIMPAT algorithm is still difficult for practical application since proposed many years ago. The improved algorithms of SIMPAT were put forward by foreign scholars, including Filtersim and DisPat. However, there are still certain defects in harmonizing the equilibrium between computational efficiency and memory space. Through deeply studying the theory and characteristics of SIMPAT algorithm, multiple-point geostatistics (MPS) algorithm based on pattern scale-down cluster was proposed. In the new algorithm, the adjacent spacing sampling method is used to perform scale-down clustering on all data patterns, so as to cluster similar data patterns into one category. Different from the one-step similarity match calculation of SIMPAT, two-step similarity contrast is adopted in the new algorithm. Firstly, a comparison is conducted on the similarities between data event and representative pattern, so as to find the most similar pattern cluster. Further, a comparison is made on the similarities between data event and the whole data pattern of such pattern cluster, thus determining the data pattern with maximum similarity. Based on 2D and 3D cases, this study also makes a comparison between the simulation computation efficiency of the new algorithm and the traditional MPS, such as SIMPAT, Snesim, Filtersim and DisPat under the same parameter conditions. The results indicate that the new algorithm can greatly improve the computational efficiency of pattern-based MPS algorithm on the basis of simulation quality assurance and also save memory spaces.