Productivity Simulation Of Superimposed Fractured Coal Measure Reservoir Based On Multi-scale Convolutional Neural Network

Liupanshui area in Guizhou has rich coalbed methane resources,and its gas-bearing coal seams are usually in multi-layer superposition distribution with thin single-layer thickness and small interlayer spacing.A multi-coal seam co-production technology using single well on such superimposed reservoirs has been developed to significantly save costs and improve gas production efficiency.The superimposed reservoirs usually have obvious differences in formation properties(reservoir pressure,fracture morphology,etc.)along vertical direction.These differences may induce vertical interference of gas flow in adjacent reservoirs and greatly affects the early production.However,no clear understanding has been available for the influence of reservoir characteristics on the interference,especially the influence of fracture morphology on interlayer interference and gas flow.It is necessary to investigate the influence of reservoir characteristics on interlayer interference in heterogeneous reservoirs with large-scale fracture network.In order to reduce the high computational cost and long time in the discrete fracture network(DFN)model,numerical simulation is widely used to calculate the coarse grid permeability of fracture network,so as to implicitly express the discrete fractures in the model and reduce computational cost.This equivalence method is called upscaling method.The calculation of equivalent permeability of large-scale DFN model in numerical simulations is also time-consuming and requires strong computational sources which are difficult to be satisfied.In order to consider the influence of fracture network on gas flow in large-scale multi-coal seam co-production and study the relationship between reservoir characteristics and interlayer interference,this thesis firstly proposes a fracture network permeability prediction model based on multi-scale convolutional neural network(Ms Net).This Ms Net model with upscaling algorithm can greatly reduce the requirement for numerical simulations and ensures prediction accuracy.It thus largely reduces simulation time,accelerates the calculation speed for upscaling,and enhances calculation efficiency.Secondly,the governing equations for fluid-solid couplings in coalbed methane production are deduced based on the dual-porosity medium assumption,and a numerical model of two-layer coalbed methane co-production is established.Thirdly,the permeability characteristics of the discrete fracture network are implicitly expressed in the numerical model of two-layer coalbed methane co-production with the Ms Net upscaling algorithm.Finally,interference coefficients of reservoir coproduction are defined to quantify the degree of interlayer interference.The influences of reservoir property difference in reservoir pressure,reservoir permeability,fracture dip angle and fracture density on commingled production capacity and gas flow are explored.Through these studies,following main results are obtained:(1)The proposed lightweight convolutional structure Ms Net can obtain a simplified equivalent DFN model and has scusceesfuly simulated the influence of fracture network on gas flow with faster solution efficiency.This Ms Net model combines the advantages of multi-scale convolution,group convolution and residual connection and its Ms Net-8-4 model performs better than the current mainstream convolutional neural network in terms of convergence speed and prediction accuracy.The Ms Net-8-4 upscaling method successfully accelerates the simulation on the gas flow in fractured coal seams with implicit expression of large-scale fracture network.(2)A numerical model of two-layer coalbed methane co-production with implicit fracture expression is established for the first time.It combined the Ms Net upscaling method with the fluid-solid coupling theory of dual-porosity media,The model can clearly describe the characteristics of gas flow under the influence of fracture network,and has successfully analyzed the influence of reservoir characteristics difference on co-production and gas flow.(3)Bigger reservoir pressure difference has greater impact on the co-production of a superimposed reservoir.Under the action of reservoir pressure difference,the gas flow in high pressure reservoir will affect the gas flow in low pressure reservoir at the interlayer and wellbore,and affect the production of low pressure reservoir through wellbore backflow and interlayer crossflow.(4)In the single-phase flow model with constant bottomhole pressure,the difference of matrix permeability,fracture dip angle and fracture density can promote gas co-production.This promotion effect is mainly caused by interlayer crossflow and decays with production time.Compared with the difference of matrix permeability,the difference of fracture density and fracture dip angle has a stronger influence on gas co-production.The wellbore interference is more obvious in the early stage.