基于核向量机的油藏历史拟合代理模型研究

项目名称： 基于核向量机的油藏历史拟合代理模型研究

项目编号： No.40872087

项目类型： 面上项目

立项/批准年度： 2009

项目学科： 金属学与金属工艺

项目作者： 程国建

作者单位： 西安石油大学

项目金额： 30万元

中文摘要： 在油田的规划和开发过程中，油藏建模是非常关键的步骤。要使所建立的模型能够用于进行生产预测，就必须使该模型与历史生产记录数据相吻合，这就是油藏历史拟合。历史拟合需要对地下流体的流动状态进行计算机模拟，其劳动密集、计算量巨大、耗费时间冗长。实际上只能进行少量的油藏模拟，而据此得到的历史拟合结果往往不完整，这就导致生产预测具有很高的不确定性。为了改善历史拟合，提高预测的可靠性，本项目将研究使用核向量机（Core Vector Machine，CVM）来构建油藏模拟器的二级代理模型。一级CVM代理充当油藏模型的"好、坏"分类器，二级CVM代理充当生产预测机。CVM代理的优势在于对高维特征、巨量样本的处理能力，其成本低廉、敏捷快速，借此可在相对较短的时段内对成千上万个油藏模型进行筛选与评价。这将使历史拟合的结果将更富含决策信息，而且据此进行的生产预测将比那些建立在小规模基础上的油藏模拟结果更为可靠。

中文关键词： 油藏模拟；历史拟合；生产预测；计算智能；核向量机

英文摘要： Reservoir modeling is a very critical step in the planning and development process of the oil fields. To make the model can be used for production forecasts, it is necessary to make the model consistent with the historical production data. This is the reservoir history matching. History matching requires computer simulation of subsurface fluid flow state. It is labor-intensive and a huge amount of computing time-consuming. Actually, we can only implement a small amount of reservoir simulation and the corresponding results of reservoir history matching were often incomplete, which lead to the oil production forecast has a high uncertainty. In order to improve the history matching, increase the reliability of the oil forecast, the project will study a class of new machine learning algorithm, CVM (Core Vector Machine), to build a two-level proxy model of the reservoir simulator. The first proxy acts as a reservoir classifier to identify which reservoir model is good or bad. The second proxy acts as a oil production forecast. The main advantage of CVM proxy lies in its high-dimensional and massive processing capacity for reservoir data. It is also low-cost, agile and has rapid screening and evaluation for thousands of reservoir models in a relatively short period. This will enable history matching results will has more valuable information for decision-making, and accordingly the oil production forecasting will be more reliable than those built on the basis of small-scale reservoir simulation results.

英文关键词： Reservoir Simulation;History Matching;Production Forecasing;Computational Intelligence;Core Vector Machines