A Generalized Bayesian Framework For Uncertainty Quantification Of A Buried-hill Fractured Reservoir

Naturally fractured oil and gas reservoirs play an increasingly important role in the exploration and development of oil and gas reservoirs.Buried-hill fractured oil reservoirs are an important type of fractured oil reservoirs,which are developed in Bohai Bay Basin and North China Oilfields etc.The existence of natural fractures greatly enhance the heterogeneity of oil and gas reservoirs and greatly change the flow properties and flow direction of fluids.Therefore,the research of fractures is very important for the effective development of buried-hill fractured reservoirs.However,natural fractures have a large span of scales.Different scales of fractures have different impacts on the reservoir,and the detection methods of fractures are also different,resulting in a complex fracture prediction process,and comprehensive research on fractures is difficult.There are large uncertainties in the understanding of this type of reservoir and the acquisition of key parameters,such as the spatial distribution of fractures,the direction of fractures,the size of permeability,and so on.In each stage of the development of buried-hill fractured reservoirs,it is necessary to establish reservoir geological models and rely on reservoir models for numerical simulation and history matching.The accuracy of the reservoir model determines the credibility of the prediction results.For buried hill fractured reservoirs,there are large uncertainties in many parameters related to fractures,and the uncertainty of some attributes caused by the limitation of detection methods in the reservoir itself,which leads to the determination of fractured reservoirs.Performance modeling and history matching are extremely difficult,and effective reservoir prediction cannot be completed.On the other hand,conventional history matching strongly relies on the experience of the engineer,which has a certain degree of subjectivity.Another weakness of manually history matching is that it usually fits a deterministic model,which will bring ambiguity,that is,history matched model is one of multiple models that can match historical data.The probability of a single model fitted to historical data cannot be quantified,which brings certain risks to model-based decision-making.Therefore,how to quantitatively analyze the uncertainty of the modeling process of fractured and fractured reservoirs and reduce the uncertainty of the model combined with historical data is of great significance for the development of this type of reservoir.This study uses a buried-hill fractured reservoir in the Bohai Basin as a basis to quantify the uncertainty in the reservoir model,to reduce the uncertainty of key parameters,and to quantitatively evaluate the uncertainty of prediction.The research content uses Bayesian theory as the framework to conduct in-depth research on the uncertainties of reservoir structure,fracture distribution,fluid distribution,and rock physical properties in buried hill fractured reservoir.In the research process,several key issues in the quantitative analysis of fractured reservoirs were studied in depth by combining a variety of machine learning methods: 1)How to comprehensively apply data from multiple sources in the process of fracture modeling and numerical simulation,such as pre-stack Post-seismic attribute data,imaging logging,thin slices and digital cores,production performance data,and outcrop observation data,etc.;2)Systematic detection method for the validity of the prior model in Bayesian inversion;3)Priori for existing problems How to adjust the prior model when the prior is falsified;4)How to use Bayes’ theorem to obtain the posterior probability for high-dimensional and multiple distribution types of parameters;5)How to solve the large number of modeling and numerical simulations required by Bayesian inversion The problem of calculating the cost.In the research on the above issues,this research has obtained the following understandings and results:(1)A framework for the qualitative quantification of uncertainties in buried-hill fractured reservoirs has been formed.The framework is based on Bayesian inversion and uses Monte Carlo simulation,falsification and adjustment of prior models,sensitivity analysis of uncertain parameters,reservoir proxy models,and calculation of posterior probability.This work framework is not only applicable to buried hill fractured reservoirs,but also has certain reference significance for the quantitative analysis of uncertainties in other types of reservoirs.(2)A systematic method for falsification of a priori model and adjustment of a priori model has been formed.For the prior model,the idea of ??anomaly detection is used for reference,and the support vector machine learning method is the core to realize the detection of high-dimensional prior models.Aiming at different situations of pseudo priors,a priori model diagnosis and adjustment method based on gradient and global sensitivity analysis is proposed.This method can be used to adjust the prior model of fractured buried hill reservoirs.(3)Bayesian approximation calculation is used to obtain the posterior distribution of parameters.This method solves the problem of the difficulty of calculation of likelihood estimation in complex models,and makes it possible to calculate the posterior probability distribution.The proposed method can be applied to a variety of parameter types and high-dimensional uncertain parameters.(4)A regression model based on decision tree is proposed,which establishes the statistical relationship between uncertain parameters and numerical simulation results.It can effectively replace the time-consuming numerical simulation process and greatly reduce the time-consuming calculation in the process of obtaining the posterior probability.(5)Through the quantitative study of the uncertainty of the buried-hill fractured reservoir,it is found that the uncertain parameters that have a greater impact on the numerical simulation of the Jinzhou 25-1 south buried-hill reservoir are the fracture opening,permeability,dynamic conductivity,and faults.Connection and blocking conditions.