由于页岩气渗流机理复杂,赋存方式多样,压裂后对裂缝网络的精确识别和表征存在较大困难,现有方法难以准确预测页岩气井产量.为此,提出了机理—数据融合建模的思路,结合连续拟稳态假设、物质平衡方程、产量递减分析方法和递推原理,建立了物理—数据协同驱动的产量预测方法,进而以中国某区块页岩气井现场生产数据为例,对该方法的准确性、可靠性进行了测试,并与经验产量递减分析和时间序列分析方法进行了对比分析.研究结果表明:①建立的产能模型采用拟压力代替压力,采用物质平衡拟时间代替时间,弱化了产量、流压和甲烷物性变化带来的影响;②以累计产量误差最小为目标开展历史拟合,弱化了生产制度变化带来的影响,使得建立的产能模型能够自动适应流压—产量变化;③应用该方法的关键在于采气指数—物质平衡拟时间双对数图中的特征直线,若图中出现特征直线,则可以开展产量预测,反之,则不能预测.结论认为:①建立的产量预测方法将不稳定流动问题转化为拟稳态流动问题求解,简化了对储层非均质性的描述,避开了裂缝网络精确识别和定量表征的难题,计算效率高,可解释性强;②生产数据测试结果表明该产量预测方法精度高,长期预测结果稳定,并优于Logistic Growth Model、Duong和Stretched-Exponential Production Decline经验产量递减分析方法,也优于非线性自回归神经网络、长短记忆神经网络时间序列分析方法.
Physics-informed data-driven shale gas well production prediction method
Shale gas is characterized by complex flow mechanisms and diverse occurrence modes,which brings challenges to the precise identification and characterization of fracture networks after fracturing and makes it difficult for existing methods to predict the production rate of shale gas wells accurately.In this regard,this paper puts forward an idea of mechanism-data fusion modeling,and then combined with the assumption of continuous pseudo-steady state,the equation of material balance,the production decline analysis method and the recursion principle,establishes a physics-informed data-driven production prediction method.Finally,with the field production data of a shale gas well in a certain block of China as an example,the accuracy and reliability of this method is tested and compared by empirical production decline analysis and time series analysis methods.The following research results are obtained.First,the established productivity model replaces the pressure with the pseudo-pressure and the time with the material balance pseudo-time,which weakens the influences of production rate changes,flowing pressure and methane physical property.Second,history matching is carried out with the minimization of cumulative production errors as the objective,which weakens the influence of production system changes,and consequently,the established productivity model can adapt to the flowing pressure-production change automatically.Third,the key to the application of this method is the characteristic line in the production index-material balance pseudo-time bilogarithmic diagram.If a characteristic line occurs in the diagram,production prediction can be carried out,otherwise,the prediction cannot be performed.In conclusion,the established production prediction method transforms the issue of unsteady flow into the solution of pseudo-steady state flow,which simplifies the description of reservoir heterogeneity and avoids the challenges to precise identification and quantitative characterization of fracture networks,so as to achieve high calculation efficiency and strong interpretability.What's more,the test results of the production data indicate that this production prediction method has a high accuracy,ensures stable long-term prediction results,and outperforms the empirical production decline analysis methods like Logistic Growth Model,Duong and Stretched-Exponential Production Decline,and the time-series analysis methods like Nonlinear Auto-Regressive Neural Network and Long Short-Term Memory Neural Network.
Shale gas wellProduction predictionPhysics-informed data-drivenArtificial intelligenceDynamic drainage areaProduction decline analysis
万方数据
维普
