基于知识驱动图版约束的致密砂岩气储层测井参数智能预测

Intelligent prediction of logging parameters of tight sandstone gas reservoirs based on knowledge-driven chart constraints

中国致密砂岩气资源潜力巨大,是天然气增储上产的重要对象,但致密砂岩储层空间类型多样,纵横向变化大,“四性”关系复杂,测井系列多样,测井项目少,常规测井技术评价致密储层参数难度大、效率低。为此,以四川盆地金秋、天府气田致密气为对象,构建构造区块—油气田—油气藏—测井解释图版主线,形成了致密砂岩气储层测井参数解释知识图谱,并通过神经网络算法对样本数据进行处理并约束模型结果,建立了图版约束的人工智能储层测井参数预测模型,实现了专家经验与数据双向驱动的储层测井参数智能预测。研究结果表明:(1)新智能模型融入了专家经验图版信息,且构建了专家经验与数据双向驱动的智能参数预测方法,极大地提升了模型对测井领域知识的理解能力和实践能力;(2)基于常规测井曲线,通过特征处理实现多维特征的挖掘,衍生出新曲线,与常规曲线一起作为输入进行模型强化训练,有助于提高解释模型的准确率;(3)实际应用结果表明,采用知识驱动图版约束的致密砂岩气储层参数智能预测方法计算的孔隙度和渗透率与岩心分析孔隙度及渗透率之间的误差分别为7.9%和15%,计算的含水饱和度与密闭取心饱和度之间的误差仅为5%。结论认为,基于知识驱动图版约束的致密砂岩气储层参数智能预测技术可以解决老井人工评价工作量大,测井解释标准不统一的问题,并可实现快速高效测井智能评价及潜力优选,将有力地推动了人工智能在测井领域的深度应用。

Tight sandstone gas resources in China have great potential and are important options of increasing gas reserves and production.However,tight sandstone reservoirs are characterized by diverse space types,significant variation in lateral and vertical directions,and complicated relations among four kinds of properties.Such reservoirs need to be logged with a variety of tools,but have not been logged sufficiently.Moreover,the application of conventional logging techniques in evaluating tight reservoir parameters is challenging and insufficient.Taking the tight gas of Jinqiu and Tianfu gas fields as an example,this paper constructs a main line of structural block-oil and gas field-oil and gas reservoir-logging interpretation chart,and establishes a knowledge graph of reservoir logging parameter interpretation of tight sandstone gas.In addition,the neural network algorithm is applied to process the sample data and constrain the model results.In this way,a chart-constrained artificial intelligence prediction model of reservoir logging parameters is established,and the intelligent prediction of reservoir logging parameters under the two-way driving of expert experience and data is realized.And the following research results are obtained.First,expert experience chart information is incorporated in the new intelligent model,and an intelligent parameter prediction method under the two-way driving of expert experience and data is established.They greatly enhance the new model's knowledge understanding and application capabilities in the field of well logging.Second,based on conventional logging curves,multi-dimensional features are mined through feature processing,and new curves are derived,which together with conventional curves are input to enhance model training,so as to improve the accuracy of the interpretation model.Third,the actual application results show that the errors between the porosity and permeability calculated by the intelligent reservoir parameter prediction method of tight sandstone gas under the constraint of knowledge-driven chart and those derived from core analysis are 7.9%and 15%respectively,and the error between the calculated water saturation and the closed coring saturation is only 5%.In conclusion,the intelligent reservoir parameter prediction technology for tight sandstone gas reservoirs can solve such problems as heavy manual evaluatiob of old wells and inconsisitent logging interpretation standards.And it can realize fast and efficient intelligent logging evaluation and optimal potential selection,which will powerfully promote the deep application of artificial intelligence in the field of well logging.