The Study On Hydraulic Fracture Propagation Of Sand-mudstone Interbeds Based On Machine Learning

There are abundant oil and gas resources in the interbed of sand-mudstone.However,there are large differences in lithology,staggered distribution of reservoir compartments and poor homogeneity of matrix,which leads to insufficient longitudinal productivity of reservoirs in actual fracturing production.The vertical cross-layer behavior of hydraulic fractures is influenced by many factors.Predicting the cross-layer behavior of hydraulic fractures in sand-mudstone interbeds is a key issue for rational design of construction scheme and improvement of engineering efficiency.Therefore,this thesis combines theoretical,experimental and numerical simulation methods to study the vertical propagation characteristics of hydraulic fractures,to clarify the main control factors of hydraulic fractures in sand-mudstone interbeds,and to establish a prediction model of hydraulic fractures propagation behavior by machine learning,which provides effective theoretical support and optimization tools for the vertical transformation of sandmudstone interbeds.The main research work and achievements are as follows:(1)Based on the hydraulic fracture initiation criterion and the vertical propagation model of hydraulic fractures in thin interbeds,the main control factors of hydraulic fractures propagation in sand-mudstone interbeds are defined,that is,the attributes of neurons in machine learning input layer,including: reservoir thickness,fracture pressure,fracture toughness of strata and stress difference between layers,and thick reservoir,high fracture pressure,low fracture toughness and stress difference between layers are beneficial to fracture propagation through layers.(2)Through the hydraulic fracturing physical model experiment of sand-mudstone layered specimens,the main control factors of vertical fracture propagation are sorted by comparative analysis,and the injecting rate has the greatest influence,followed by the strength of cemented surface,followed by the thickness of interlayer and the smallest is viscosity.Among them,high injecting rate and low viscosity fracturing fluid promote the fracture height extension,while weak cementation surface and thick interlayer inhibit the vertical propagation of fractures.(3)Based on the cohesive element,the intersection model of hydraulic fracture and interface fracture is established,and the grey relational analysis method is used to judge the main control factors of the characteristics of hydraulic fracture height and width in sandstone-mudstone interbeds.The order of influence degree is as follows: the ratio of elastic modulus of sandstone,injecting rate,formation stress difference,rock tensile strength and fracturing fluid viscosity,and the chart of critical conditions for hydraulic fracture penetration is established.(4)Based on the research results of theory,experiment and numerical simulation,the characteristic attributes of neurons in the input layer of BP neural network are determined,and the prediction model of hydraulic fracture propagation in sand-mudstone interbeds is established.The accuracy of the model is over 90% as verified by the test set data.The model is applied to the optimization of fracturing construction scheme of two wells in the field,and the volume of fracture reconstruction is effectively increased.