【作者】 张杰；

【导师】 孟凡顺；

【作者基本信息】 中国海洋大学 ， 海洋地球物理学， 2014， 博士

【摘要】 随着全球各大油田开采时间的延续，油田的地质条件越来越复杂，油水井发生套管损坏的频率越来越高、容易造成停产甚至报废井的事故，给油田带来巨大的经济损失。套管损坏问题是全球范围内急需解决的难题，各大油田都存在套管损坏问题。因此，寻找合理的预测方法对套管状况进行及时的预测并采取合理的预防措施，对延长油水井的工作寿命、降低损失从而提高生产效率具有重要意义。套管损坏问题是一个影响因素多、机理复杂的系统，其具有复杂性、不确定性、模糊性、定性分析难以定量化等特点。本文结合海拉尔油田的实际生产数据，在现有套损井数据基础上，通过数据资料统计分析，研究各个区块套管损坏的影响因素，获得各区块套管损坏的主控因素；研究表明，海拉尔三大研究区块的套管损坏是多种因素共同影响作用的结果，其核心在于断层导水、泥岩吸水膨胀、储层物性较差、憋压严重、油压过高、粘土矿物含量高、高含水、增产措施等因素导致油水井套管发生了不同程度的损坏。因此，本文从系统的角度，介绍了套管损坏的基本研究方法、常用的预测方法，通过对比分析选择了模糊数学的综合评价方法和人工神经网络方法进行研究，模糊数学综合评价方法适用于解决因素复杂、评价标准模糊、数据定性分析难以定量化等问题，人工神经网络方法适用于解决数据之间存在非线性关系、个别数据有误差等问题。然后结合海拉尔油田实际生产数据，从不同的研究区块入手，全面分析了三大区块开发过程中套管损坏的影响因素。针对套管损坏影响因素和评价标准具有模糊性、难以定量化分析的局限性，提出了基于层次分析法确定权重的模糊综合评价预测方法，并建立了相应的套管损坏预测模型，在套管损坏定性分析的基础上实现了定量分析，对研究区块已知套损井进行了实例计算和验证，表明模糊综合评价模型的评判结果具有较高的可信度；针对套管损坏影响因素数据存在误差值、难以建立线性函数关系的局限性，提出了基于BP神经网络方法的套管损坏预测方法，并建立了相应的套管损坏预测模型，学习训练样本的数据结果显示模型构建合理，对样本的预测符合实际情况。在理论研究的基础上，以MapInfo Professional、Map Basic为平台，开发了实用可靠的套管损坏预测分析系统，并结合海拉尔油田三大区块实际生产数据进行了预测。预测结果表明，使用模糊数学预测系统对套管状况进行预测具有使用方便、参数选择适应性高、界值确定灵活性大等特点，使用人工神经网络预测系统对套管状况进行预测具有学习训练样本选择智能化、泛化能力强的特点，为油田套管损坏预测研究提供了新的途径，具有较好的应用前景。更多还原

【Abstract】 With the world’s major oilfield has been exploited more and more complex oilfieldgeological conditions cause the Increasingly high Frequency of the casing damage of Oilwells. It takes huge economic losses to oilfield by causing accidents and making wellsabandoned. So, It’s great significance to find a reasonable prediction method for predictingcasing situation, the extension of the working time of oil wells and reduce losses.Casing damage is a systems with many affecting factors and complex mechanism, whichhas the characteristics like complexity, uncertainty, ambiguity and qualitative analysis isdifficult to quantify. In this paper, through statistical analysis of data from Hailar oilfieldproduction and the existing data on the basis of the well casing damage. research InfluencingFactors each block to find the major factor. Research shows that the casing damage of Hailarblock is a variety of factors influence the outcome such as water guiding tomographic,swelling shale, poor reservoir properties, serious pressure, the high oil pressure, the highcontent of clay minerals, high watery.Therefore, this paper introduces the basic casing damage research methods andcommonly used prediction methods withsystem. Through comparative analysis of FuzzyMath and Artificial Neural Networks, find Fuzzy Math is suitable for solve complex factors,evaluation criteria fuzzy and difficult to qualitative analysis. Artificial Neural Networks issuitable for solve the nonlinear relationship between the data and Individual data errors. Thencombine Hailar oilfield production data, studies different blocks, comprehensive analysis ofthe factors affecting the development process of Casing damage. Direct factors of Casingdamage and limitations of the evaluation criteria are fuzzy and difficult quantitative analysis.Propose a fuzzy comprehensive evaluation to determine the weight of the forecasting methodbased on AHP and the establishment of appropriate Casing damage prediction model, achievea quantitative analysis on the basis of the qualitative analysis of the casing failure, calculationand verification examples to the well known the Casing damage. Evaluation results show that fuzzy comprehensive evaluation model with high credibility;Direct factors of Casing damagehave limitation of error value and dfficult to establish a linear function, proposed Casingdamage prediction method based on BP neural network method, and the establishment ofappropriate casing damage prediction model. Data show the results of learning and trainingsamples to construct a reasonable model, sample forecasts comply with the actual situation.On the basis of theoretical research, in MapInfo Professional, Map Basic platform,developed a practical and reliable Casing damage prediction and analysis system,combinedwith actual production data of three blocks Hailar oilfield to predict. The prediction resultsshow that, using Fuzzy Math forecasting system, it has Feature as easy to use, highadaptability parameter selection, Flexibility to determine the critical value. Using ArtificialNeural Networks forecasting system has feature as sample Selection intelligent learning andtraining, strong generalization ability. Provides a new way for prediction of oilfield Casingdamage and has good prospects.更多还原