Quantitative Characterization Of Remaining Oil In Shahejie Formation Of Gao30 Fault Block In North China Oilfield

The Gao30 fault block in the North China Oilfield is dominated by multi-layer sandstone reservoirs.The oil and water distribution of the reservoir gets complicated during the later high water cut periods,which increases the difficulty to accurately describe the existence mode,distribution pattern,control factors and potential scale of the remaining oil.This paper first applies machine learning to predict reservoir parameters and improve the prediction accuracy of porosity and permeability,and further uses neural network technology to automatically identify sedimentary microfacies and build 3D facies model,which not only simplify manual operations but also improve the accuracy of sand model.On the basis of this,numerical simulation is used to quantify the remaining oil distribution and eventually provide a solid basis for the petroleum development plan.The overall sedimentary environment of Es₂ reservoir is a shallow-water delta front,it can be divided into four kinds of sedimentary microfacies,including subaqueous distributary channel,interdistributary bay,overflow bank,and a small amount of estuarine bars.In the later stage of development,the well pattern is relatively dense,therefore,the single-well sedimentary microfacies identification and processing workload are large.This paper uses neural network image recognition technology to automatically identify the single-well facies and develops the automatic identify method of single-well sedimentary microfacies identification with accuracy of 95%.Based on core and logs calibration,machine learning algorithms such as Random Forest,Gradient Boosting Tree,and XGBoost are applied in lithology identification and reservoir parameter prediction.The results indicate that in terms of lithology identification,XGBoost algorithm generates the lowest error,with accuracy reaches up to 94%.As for parameter prediction,the porosity prediction error of machine learning algorithms is about 1.2%lower than the traditional linear regression,and the error distribution range is reduced by 30%～50%;The permeability prediction error is about 1m D lower and the distribution range is reduced by 20%～40%.Overall,the machine learning algorithm improves the recognition accuracy while reduces the error in the prediction.The technology of Generative Adversarial Network is applied to the actual oilfield3D geological modeling.The coordinates are used as condition information in the training process of the neural network,which improves the simulation efficiency and accuracy,and also randomly generates a 3D model that matches well data.Controlled by the distribution pattern of sedimentary facies,the reservoir petrophysical models are established using Sequential Indication Simulation,and followed by the reservoir numerical simulation with the fitting agreement rate reaches to 92%.Based on the results of comprehensive analysis of reservoir geological characteristics and numerical simulation results,this paper analysis the distribution characteristics and influencing factors of the remaining oil.Results show that the distribution of remaining oil is mainly affected by four aspects,including water injection well pattern,structural morphology,edge and bottom water and heterogeneity.The remaining reserves of the Es₂ reservoir in the Gao30 fault block are 2.185 million tons,with an average recovery percent of 25.3%.The remaining reserves of Es₂₂ and Es_23,the main oil-bearing formations,have around1.828 million tons,accounting for 83.7%of the total remaining reserves within the Es₂reservoir.