Study On Optimization Of Injection-production Parameters Of Waterflooding Reservoir Based On Intelligent Algorithm

In the process of water flooding reservoir development,it has always been a difficult problem to determine the injection-production parameters of injection wells and oil production wells.Different parameter combinations will have a great impact on the oil production,water breakthrough time and oil recovery.In the heterogeneous reservoirs,the influence of injectionproduction parameters is particularly prominent.Traditional optimization methods rely on numerical simulation,which requires calculation and comparison of many different schemes.This process often takes a particularly large amount of time.Therefore,this paper introduces intelligent algorithms based on machine learning,uses deep learning networks as prediction models,and stochastic algorithms as optimization methods,to study the inherent problems in the process of waterflooding development prediction and optimization.The main results of the thesis research work include:(1)This study discusses the main influencing factors of waterflooding development(including reservoir properties,fluid properties and production system)and the main evaluation indicators(development indicators and economic indicators)of waterflooding development.At the same time,it discusses and analyzes the limitations of traditional injection-production parameter optimization methods,such as too large calculation amount and limited search options,and the concept of proxy model is introduced to solve this type of problem.(2)Based on the convolutional neural network,the concept of transposed convolution is introduced,and the TCNN waterflooding proxy model is established using Python.The model regards different injection-production parameters and production time as high-level features of the oil saturation and pressure distribution field in the entire region and restores the abstract features of the picture to the original image through convolution and transposed convolution.In this way,the oil saturation and pressure distribution field image can be restored,and the production performance of waterflooding can be predicted.(3)The study discussed the impact of data set processing methods(including experimental design method,data preprocessing,data standardization)and parameter selection(including loss function,learning rate,sample size)on the accuracy of the model during the training of the TCNN waterflooding proxy model.A 60×60 example reservoir was used to verify the accuracy and efficiency of this method for waterflooding production prediction.Compared with numerical simulation software(MRST),the TCNN proxy model can save more than 95% of the calculation time,and its error in predicting water and oil production is less than 1.8%.(4)This paper establishes a mathematical model of waterflooding optimization and introduces three non-gradient optimization algorithms(including particle swarm algorithm,genetic algorithm,differential evolution algorithm)combined with TCNN proxy model to optimize injection-production parameters for waterflooding reservoir examples design,which verifies the applicability of this method to waterflooding optimization problems.Compared with the basic plan,the total water injection volume of the optimized plan is only 59.74% of the original plan,while the NPV is increased by about 31.84%.The optimal plan not only improves the economic benefits,but also delays the water breakthrough time in the study area.The optimized plan has significantly improved the development effect of the research area.