基于交叉检验法的地质预测模型优选  

Selection and assessment of geological prediction models based on cross validation

作　　者：宋永康[1] 舒晓 王兵杰[3,4] 

机构地区：[1]中国石油辽河油田分公司,辽宁盘锦124109 [2]中国石油大学(华东)地球科学与技术学院,山东青岛266580 [3]南京大学内生金属矿床成矿机制研究国家重点实验室,江苏南京210093 [4]南京大学地球科学与工程学院,江苏南京210093

出　　处：《石油化工应用》2013年第12期65-69,88,共6页Petrochemical Industry Application

摘　　要：在地质建模领域中,有效地从多种建模方法和变差函数组合中选择出最优方案作为地质预测模型是一个亟需解决的问题。传统方法往往采用定性的目视化对比选择法,但其滞后性大、效率低且优选的有效性无法保证。因此,本文提出应用交叉检验法进行定量的预测模型优选及评估,同时简要分析了建模过程中实验变差函数计算及拟合需要注意的问题。应用该方法进行了实际工区气井初期产能值预测模型的优选,加密井验证结果表明:应用交叉验证法优选得到的地质预测模型有着最准确的气井初始产能值估计结果,证实了该优选方法具有良好的实际应用效果。How to obtain the most appropriate geological prediction model effectively and accurately which can reflect the spatial heterogeneity and continuity are often severe challenges researchers face in geostatistical modeling.Conventional method like visual-based comparative method couldn＇t choose the optimal model effectively and have low efficiency, high hysteresis quality and poor accuracy.This paper proposes a way to identify the optimal prediction model using cross validation and gives some practical advice to compute and model empirical variograms. Practical validation results show application of cross validation can choose the most optimal geological prediction model from multiple candidate models and get a more accurate estimate result which show its effectiveness and correctness.

关 键 词：模型优选 交叉检验 变差函数 克里格方法 去一法 去从聚效应 比例效应 

分 类 号：P624[天文地球—地质矿产勘探]