【作者】 田景文；

【导师】 李国斌；

【作者基本信息】 哈尔滨工程大学 ， 控制理论与控制工程， 2001， 博士

【摘要】 石油的产生有着不可重复性，不可实验性的特点。人们对地下油藏的认识存在很多的疑问，由于某些不正确的认识会带来巨额生产资金消耗。研究清晰地再现地下油藏构造形态和准确预测油藏参数的方法，对石油的生产有重大的实际意义，可以提高生产效益，为国家节约巨额的资金。 本文以大庆油田和辽河油田的科技攻关项目为背景，以准确、清晰地再现地下油藏的构造形态和油藏参数空间展布为目标，将计算机仿真技术和现代预测理论与石油生产相结合，围绕对地下油藏的仿真和油藏参数预测进行一系列的研究工作。主要内容如下： 1．在分析地震信号特性的基础上，研究设计了一种自适应数据采集系统，该系统采用智能放大器，可以完成对地震数据大动态范围的高分辨率采集，而且该采集系统设计为具有2⁰～2¹⁵共16个可选数字增益台阶，可根据对输入信号的预测结果，自动选择一个最佳增益对输入信号进行放大，系统具有自动零漂补偿和自动定标的能力，该系统可以提高数据采集的分辨率，扩大采集信号动态范围，为准确再现地下油藏构造形态，提供原始的数据。 2．深入研究了小波神经网络理论，并将地震数据剖面看成是二维数字图像。提出将小波神经网络边缘检测算法应用到地震剖面图像的特殊处理上，提高对油藏的分辨率，从而寻找地下小的含油构造和薄的储油层，使计算机能更准确地再现地下油藏的构造形态，提高勘探的精确度。并用该方法对实际资料进行了处理，取得了较好的效果。 3．对三维地震数据体的数据结构进行了分析，建立地下油藏的三维可视化仿真模型。利用计算机图形图象处理技术，研究对地下油藏仿真模型的切片显示的方法，其中①可完成多层水平片、可使人们在纵向空间对油藏有清晰的认识，②完成多方向、多角度对油藏的垂直切片，使人们对油藏的整体构造有清晰的认识。③完成对地下油藏在不同层位的剥层显示，使人们对地下油藏各层位的空间构造形态有非常直观清晰的了解。通过以上三种方法可以清晰地显示地下油藏的构造形态，文中以某油田的一个地区的地下油藏为例给出切片和剥层的实际仿真图。 4．对克里金估计技术、贝叶斯-克里金估计技术进行了深入分析，针对以往克里金估计技术预测油藏参数的不足，提出了贝叶斯-克里金油藏参数 哈尔滨工程大学博士学位论文一预测方法。由于克里金估计技术是利用单一的精度较高的井点数据对油藏参数进行预测，当井较少时，该方法受到限制。本文提出贝叶斯－克里金估计技术可以将少量精度高的井点数据和大量精度低的地震数据有机结合，充分发挥各种数据的作用，完成对大范围的油藏参数的预测，提高预测精度。通过实际资料验证，效果显著。 5．深人研究了遗传算法的理论，针对遗传算法和BP网的不足及薄互油藏参数预测的特点，提出了快速、高精度遗传算法人工神经网络的薄互油藏参数预测新方法。在方法研究过程中，注重对遗传算子和控制参数改进的研究。充分利用现有的样本数据，发挥神经网络的非线性逼近特性和遗传算法的全局搜索能力，对薄互油藏参数进行快速、准确的预测。用该方法对实际数据进行了预测，取得了较好的效果。 6．在深入研究模拟退火算法、Powell算法的理论基础上，针对模拟退火算法收敛速度慢问题，提出了两种快速模拟退火组合优化算法，即最佳保留模拟退火与Powell有机结合的FPSA和最佳保留模拟退火与遗传算法有机结合的F．GSA，并将它们与BP网络相结合应用到薄互油藏参数的预测中。对于不同地区，由于地区性差异，当根据实际情况，网络隐层和输入神经元较少时，将FPSA与BP网结合，完成参数预测。当网络隐层和输入神经元较多时，Powell算法起不到加快网络收敛速度的作用，这时将FGSA与BP网结合进行薄互油藏参数预测。文中给出了FPSA组合优化算法人工神经网络薄互油藏参数预测实例和各种算法的仿真结果。验证了F－PSA和F－GSA这两种组合优化算法神经网络对油藏参数预测精度较高，算法收敛较快。 本文研究成果和结论对石油的勘探、开发和科研具有重要的指导意义与实用价值。更多还原

【Abstract】 Petroleum cannot be reformed by experiment,and man cannot made it repeatedly underground. Man have some unclear cognition about underground oil reservoir,it make waste a lot of money every year. So it is very important that man realize the structure and parameter of oil reservoir clearly. It can heighten production efficiency and save a lot of money .In this thesis,some scientific research item of Daqing and Liaohe oil field are used as the background. And it is used as object that the modality of oil reservoir is reappeared clearly. We make three-dimensional visual emulation to the underground oil reservoir . Also we study the methods for accurate fast prediction the parameters of oil reservoir. The main contents are as follows:1. One kind of intelligent amplifier is studied. It have 16 grades numeric gain,and compensate zero-drift automatically,and prediction input signal then adjust gain automatically. One kind of adaptive data acquisition system is constituted using computer and intelligent amplifier. The distinguishability of data acquisition system is elevated. The circuits of the are designed. This system can provide good quality data for the follow work.2.The method of marginal checking using wavelet neural network is used to process the seismic data for increasing distinguishability to Recognition the thin interbedded oil reservoir .Small structure of oil reservoir and thin interbedded are find . Oil reservoir can be reappeared on computer clearly.3. Three-dimensional emulation model is established using seismic data. The methods of displaying slice about the three-dimensional emulation model are studied. They are follow three methods.the method to make multilayer horizontal slice;the method to make multidirection and multiangle slice the method to make single oil layer display in three space . By above three methods the oil engineer can apprehend the underground oil reservoir structure very clearly. In this thesis,an oil reservoir is selected as example and some typical images are offered.4. Krijing estimation and Bayes-Krijing estimation technique are studied.Contraposed disadvantage of Krijing estimation on prediction oil reservoir,Bayes-Krijing estimation technique are studied to predict oil reservoir parameter Because Krijing estimation depend on well data to predict oil reservoir,while the wells is few,it is limited .Here Bayes-Krijing estimation technique is studied to predict oil reservoir. The seismic data and the well data are combined in Bayes-Kerijing estimation to predict oil reservoir parameter. By this way,we can get better estimation result of oil reservoir parameter.5. In allusion to the disadvantage of genetic algorithm and BP neural network,the genetic algorithm BP neural network to predict oil reservoir parameters is studied . In this thesis the genetic operator and control parameter are ameliorated. Good nonlinear approach ability of neural network and searching optimization solution ability of genetic algorithm are combined to predict oil reservoir parameters . Actual application result appear that this algorithm predictoil reservoir is accurate and fast.6. Based on studying simulated annealing algorithm and Powell algorithm,in allusion to simulated annealing algorithm convergence slow,the two fast simulated annealing combination optimization algorithm are offered. One is optimal reservation simulated annealing and Powell combination algorithm (F-PSA). The other one is optimal reservation simulated annealing and genetic algorithm optimal combination algorithm(F-GSA).And these two optimal combination algorithm are combined with BP neural network to predict oil reservoir parameters. In actual application the data of different area have different characteristic,when the hidden layers and input nerve centers are less,we use F-PSA combining with BP to predict oil reservoir parameters;Otherwise we use F-GSA combining with BP to predict oil reservoir parameters .In this thesis some emulation results and actual application results are offered . Those result更多还原