【作者】 钟仪华；

【导师】 杜志敏； 刘志斌；

【作者基本信息】 西南石油大学 ， 石油工程计算技术， 2008， 博士

【摘要】 目前我国东部各油田总体已进入特高含水开发阶段,准确的油田开发动态预测及科学的油田开发规划是这些油田合理、高效、可持续开发的基础。现有的油田开发动态预测方法有单变量开发指标的趋势预测方法和单变量开发指标基于影响因素的预测方法,但它们因在不同开发阶段,适用范围不尽相同、对预测要求的条件以及复杂程度也不同,因而各有其局限性；基于多输入多输出开发指标及影响因素的预测方法在理论和应用中还存在许多有待深入研究的问题。本文在分析研究油田开发动态预测常用方法的适用性基础上,解决了制约基于多输入多输出开发指标及影响因素的预测方法的理论和应用问题,建立了符合特高含水油田开发动态变化规律的新的开发规划动态预测方法。它包括支持向量机预测方法、基于时变系统的功能模拟预测方法、概率模拟预测方法、组合预测方法和智能预测方法。前三种方法本质上都是基于油气动态系统的功能模拟性,依据油田开发生产历史数据,利用功能模拟原理,建立开发动态预测系统的输入和输出体系,进而建立相应开发指标的预测模型。支持向量机预测重点关注系统的小样本预测问题；基于时变系统的功能模拟预测则关注具有时变系统的多步预测问题；概率模拟预测更注重系统的随机不确定性问题；组合预测解决提高预测精度的问题；智能预测侧重自适应和自主选择最合适预测方法问题。这些预测新方法的实例应用研究表明,建立的特高含水油田开发指标预测体系符合实际,提出的预测新方法和设计的算法正确可行。本文发展的适合不同油田特高含水阶段开发规划动态预测方法、确定选择预测指标和影响因素的标准及研究油田开发指标与其影响因素相关性的方法,丰富了油田开发动态预测的理论和方法,为编制和制定特高含水油田的科学开发规划方案提供了理论依据和重要技术支撑。本文主要进行了以下几方面的研究：(1)针对特高含水油藏的开发特征和开发面临的问题,分析了油田开发动态预测常用方法的适用性,比较并评价了它们遵循的机理和局限性,得到现有的常用预测方法都不太适合用于特高含水油田开发规划动态预测；指出了特高含水油田开发规划动态预测方法的发展趋势。(2)基于工程和数学的理论和方法创建了特高含水油田开发指标及影响因素相关性分析的方法。依据特高含水油田开发指标的动态变化规律特征,基于油藏工程、油田开发等理论确定出不同油藏特高含水水驱油田的可能开发指标及影响因素,利用关联分析方法对开发指标及影响因素进行相关分析；或依据油气动态系统预测的功能模拟性,从开发指标及影响因素的历史数据出发,利用基于模糊相关性分类的方法对开发指标及影响因素进行相关分析。形成了确定选择预测指标和影响因素的标准,建立了不同油藏特高含水开发阶段的开发指标及其影响因素体系。(3)针对特高含水油田开发阶段一般采集到的油田开发指标及影响因素的样本数较小,和现有方法的局部极小点、过学习和欠学习等问题,利用多目标优化理论剖析了支持向量机方法,研究了核函数和参数选择问题,提出利用开发指标时间序列特征分析的数据挖掘方法选择核函数的单输出和多输出支持向量回归机方法。(4)就常用预测方法进行多步预测的准确性或中长远预测效果较差的问题,综合神经网络模拟的拟合精度高、预测指标与影响因素联系紧密及微分模拟预测在预测过程中注重预测指标的自身变化趋势的优势,建立了具有时变特征的、预测模型中的参数随时间变化,对中长期预测有更好效果的基于时变系统的功能模拟方法。(5)研究了解决单一预测方法信息不全、方法各有缺陷的组合预测方法的原理和方法,提出了确定组合预测方法权重的改进方法。研究了智能预测方法的原理和方法,创建了油田开发指标系统的智能预测方法,实现开发指标的自动预测。(6)鉴于油田进入特高含水开发阶段,影响油田开发的地质和人为因素的随机不确定性更严重,及目前所有的开发指标预测方法未进行随机性和可能性研究的缺陷,全新地从理论和实用上探索了油田开发指标系统的不确定性预测方法和模型。(7)预测方法应用研究。设计了特高含水油田开发规划动态预测新方法的算法,开发了应用程序,并将其应用于国内某油田特高含水阶段的开发指标预测,预测结果表明本文建立的特高含水油田开发规划动态预测方法具有较大的理论和实用价值,值得进一步推广应用。更多还原

【Abstract】 Accurately dynamic prediction of oilfield development and scientific oilfield development programming are the basis of the reasonably, efficiently and continually developing of the east oilfields of China, for they presently have generally entered ultra-high water-cut development stage. The exiting methods of oilfield development dynamic forecasting are the trend forecasting method of single variable development index and the prediction method of single variable development index based on affecting factors. However, each method has its limitations, which its applied scope, prediction conditions and complexity are all not the same in the different development stages. The prediction method based on the multi-input and multi-output development indexes and their factors still need to further study in theory and application. With the applicability analysis of the methods commonly used in oilfield development dynamic prediction, new dynamic prediction methods for development programming according with the law of dynamic change of oilfield development in ultra-high water-cut stage are established in this paper, these methods solve the theory and application problems which restrict the prediction method based on the multi-input and multi-output development indexes and their factors. The new methods include support vector machine forecasting, functional simulation prediction based on time-varying system, probability simulation forecasting, combination forecasting and intelligent forecasting. The first three methods utilize the functional simulation method to establish the input-output of development dynamic predication system and the relevant prediction model of development indexes essentially based on the functional simulation of oil and gas dynamic system and the oilfield development historical data. The prediction method of support vector machine forecast system is mainly focused on prediction problems with a small of samples; the functional simulation prediction with time-varying system is focused on multi-step prediction of time-varying system; the probability simulation forecasting is more focused on random uncertainty of system; the combination forecasting solves the problem of improving forecasting precision; the intelligent forecasting emphasizes particularly on selecting adaptively and independently the most appropriate prediction method.The experimental results of these new prediction methods show that oilfield development dynamic index prediction system in ultra-high water-cut oilfield conforms to reality, and the designing algorithms are correct and feasible. It enriches the theory and method of dynamic prediction of oilfield development, and provides the theoretical basis and important technique support with making and establishing the scientific schemes of development programming for ultra-high water-cut oilfield that dynamic prediction methods of ultra-high water-cut oilfield development programming which was suitable for the different oilfield being developed, the standards of selecting prediction index and its influence factor being determined and the studying method about the correlation between the oilfield development index and its influence factors be presented in this paper. Main topics involved in the thesis are as follows:(1) With development characteristics and facing problems of ultra-high water-cut oil reservoir, this paper analyzes the applicability of common methods used in dynamic prediction of oilfield development; Compares and evaluates the laws and limitations these methods followed, and then draws the conclusion that the existing common methods are not suitable for the dynamic prediction of ultra-high water-cut oilfield development programming; and points out its development trend.(2) Based on the theory and method with engineering and mathematics, the method of correlation analysis between development index and its influence factors of ultra-high water-cut oilfield is established. According to dynamic change law of ultra-high water-cut oilfield, the possible development indexes and their influence factors of different ultra-high water-cut oilfield are first determined by the theory of reservoir engineering and oilfield development etc. Then a correlation analysis of development indexes and their influence factors is made through using relational analysis. Or according to functional simulation of oil and gas dynamic prediction system, a correlation analysis of development indexes and their influence factors is made by using fuzzy correlation classifying method starting from historical data of development indexes and their influence factors. As a result, the criteria of selecting prediction index and its influence factors are determined, and the system of development indexes and their influence factors of the different oil reservoirs in ultra-high water-cut stage are established.(3) According to the problems of the less sample numbers of development indexes and their influence factors generally collected in ultra-high water-cut oilfield stage, local minimum point, over learning and under study with existing method, after this paper analyzed the prediction method of support vector machine by utilizing multi-objective optimization and studied the kernel function and parameter selection problem, it puts forward a method for selecting kernel function of single output and multiple output support vector regression machine by using feature analysis data mining method of development index time series analysis.(4) For the accuracy problem of multi-step prediction by using common prediction method or the poor effect of middle-long term prediction, a functional analogue method based on time-varying system with the time varying characteristic, parameter of prediction model changing with time and better effect to middle-long term prediction is established by integrating the high fitting accuracy and close relation between prediction index and its influence factors of neural network simulation with the advantage of differential simulation prediction paying attention to the change trend of prediction indexes themselves.(5) After studying the principle and method of combination forecasting for solving the problems of information inadequacy of single prediction method and its limitation, this paper presents an improved method to determine the weight of combination forecasting method. Though studying the principle and method of intelligent prediction method, the intelligent forecasting method about the oilfield development index system is established to realize automatic prediction of development index.(6) Considering the more serious stochastic uncertainty of geological factors and man-made factors affecting oilfield development in ultra-high water-cut stage and the insufficiencies without studying randomness and possibility of all existing development index prediction methods, the uncertainty prediction method and model of oilfield development index system are explored theoretically and practically in this paper.(7) Study on application of forecasting method. The algorithm and the application program for new methods of dynamic prediction of ultra-high water-cut oilfield development programming are designed and developed. The development index prediction case of one oilfield of china in ultra-high water-cut stage is studied with the presented method in this paper. The prediction result shows that dynamic prediction method of ultra-high water-cut oilfield development programming established in this thesis has bigger theoretical and practical value, and it is worth of expanding range of application.更多还原