| 油田混层开采产能动态监测技术 |
| 孔枫
|
| 2007-02-10
|
| 学位授予单位 | 中国科学院地球化学研究所
|
| 学位授予地点 | 地球化学研究所
|
| 学位名称 | 博士
|
| 关键词 | 色谱指纹
支持向量机
产能分配
|
| 摘要 | 本论文在运用色谱分析研究原油全烃技术理论的基础上，通过大量的单层原油和混合原油的模拟配比实验，运用色谱烃指纹峰高比、内标法绝对定量分析等方法建立了化学模型，利用非线性神经网络方法和支持向量机建立了多层混采原油分层产能贡献数学模型，为实现油田混层开采产能动态监测提供了方法。
本论文重点对文昌13-1/2油田的混采油井产能贡献动态监测进行了研究。首先，注重了解油田的地质现状，避免因为地质因素的误判所带来的计算误差。有机相和源岩成熟度的变化决定了石油组成差异，导致了油藏单元之间色谱指纹不同，因此原油色谱指纹的变化是原油非均质性的具体表现。再次，研究针对不同混采条件建立了本区多层混采油井分层产量贡献的判别模型。对于两层合采问题的数学模拟计算，一般用线性拟合的方法。对于三层及以上多层合采问题，分层原油与合采原油烃指纹浓度之间已不是简单的线性关系。因此，需要通过建立合理的数学模型来实现对多层合采原油分层贡献的计算。目前常用的非线性的预测模型是采用神经网络，然而深入的研究表明神经网络虽然具有较强的模式预测能力，但是算法的收敛速度较慢，并且无法保证其全局最优。基于统计学习理论的支持向量机，实现了数据空间和特征空间之间的非线性映射，可有效地将数据空间中的各种非线性操作演变为特征空间中相应的线性操作，进而大大提高了非线性处理能力，并且在解决小样本，局部极小点以及高维数等模式预测问题中表现出许多特有的优势，成为近年来新的研究热点。本研究应用神经网络和支持向量机分别建立了数学模型及软件，对文昌13-1/2油田混采油井产能分配进行动态监测，计算了5口混采油井的分层产能贡献，其解释结果与生产测试结果基本一致,为该油田制定生产调整方案提供理论依据。 |
| 其他摘要 | By applying the methodology as outlined in this paper, study was carried out on dynamic monitoring of production potential-based contributions from multiple oil-producing horizons in an oil well in Wenchang 13-1/2 oilfield.
First of all, the study was focused on geological investigation of the current status of the oilfield. they were studied in order to avoid calculation errors that may de derived from misjudgment of geological factors. Variations in GC fingerprint features of crude are generally regarded as representing the inhomogeneity of the crude, this is simply because changes in maturity of organic phase or hydrocarbon source rocks have determined the differences in petroleum compositions, which will in turn result in the differences in GC fingerprint features between different oil reservoir units.
Secondly, a model dedicated to identification of production from individual horizons in an oil well consisting of multiple oil-producing horizons in the oilfield, was established under different conditions for combined exploitation of the oil well. In regard to mathematic simulation and calculation for combined exploitation from two horizons, the method of linear fitting is generally adopted. In regard to combined exploitation from three horizons or even more horizons, the fingerprinting hydrocarbon concentrations in crudes exploited from individual horizons are not linearly related to those in crudes derived from combined exploitation of multiple horizons. As a result, a realistic mathematical model has to be set up in order to perform calculation of contributions from individual horizons for combined exploitation from multiple oil-producing horizons. The most commonly used non-linear predictive model is the neural network at the moment, but an in-depth study reveals that neural network calculation shows slow speed in convergence and fails to ensure optimization of the entire calculation program, even though it proves to be strongly capable in modeled prediction. Support vector regression (SVR) based on statistical learning theory is one such method of neural network calculation. SVR can help to realize non-linear mapping between data space and feature space, and effectively transform all kinds of non-linear operations in data space into their corresponding linear operations as defined in feature space, hence dramatically enhancing the capability in non-linear processing. Moreover, SVR shows many unique advantages in providing solutions to model prediction issues such as small samples, local minimum points as well as high dimension numbers, so becoming a hot topic in recent research.
By applying the methodology as discussed above, a mathematical model was set up and a relevant software was developed in this paper. The model and software were used for dynamic monitoring of distribution of production potential of an oil well characterized by multiple oil-producing horizons in Wenchang 13-1/2 oilfield. Contributions, in terms of production potential, from individual horizons in 5 oil wells consisting of multiple oil-producing horizons were calculated, and the results of geochemical interpretation prove to be basically consistent with those from production testing. We believe that the calculation program can form a theoretical basis for stipulation of an improved production plan for this oilfield. |
| 页数 | 83
|
| 语种 | 中文
|
| 文献类型 | 学位论文
|
| 条目标识符 | http://ir.gyig.ac.cn/handle/352002/3128
|
| 专题 | 研究生_研究生_学位论文
|
推荐引用方式
GB/T 7714 |
孔枫. 油田混层开采产能动态监测技术[D]. 地球化学研究所. 中国科学院地球化学研究所,2007.
|
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。
修改评论