Research On Data-driven Modeling And Dynamic Programming Of Alkali-surfactant-polymer Flooding

After the first and second period of oil development,the oilfields development enter into the third period.In the period,the moisture content becomes higher and higher,the oil production becomes less and less,so it is very important to research the new technology to enhance oil recovery.ASP flooding is a new technology which received good testing results in many oilfields,but it has some shortcomings,including high price,long injection period,slow effect,and complex models.In order to enhance the overall economic benefits,it is of vital important to research the ASP flooding model to draw up scientific and reasonable development plan.In this paper,we first propose interactive multiple-model predicting method to build moisture content prediction model,then propose the spatiotemporal decomposition model of ASP flooding based the Gaussian processes,finally use iterative dynamic programming method to solve the optimal control problem of ASP flooding.Aiming at the shortcomings of KH algorithm in complex optimization problems,we make some improvements.A good-point KH algorithm based adaptive Cauchy mutation is proposed.Firstly,the good-point set is utilized to initialize the krill population and improve the ergodicity of initial population.Secondly,the speed factor is updated according to the changes of krill population.Finally,adaptive Cauchy mutation is adopted to enhance the algorithm's ability of jumping out of local optimum.The simulation results on standard test functions and chemical processes show the effectiveness of the proposed algorithm.Aiming at the complex nonlinear characteristic,interactive multiple-model predicting method is proposed to predict the moisture content of ASP flooding.Gaussian process has few optimization parameters and can calculate the uncertainty of the output.Regular extreme learning machine has fast learning ability and nice generalization performance.And the two models are both used to build the prediction model of moisture content.Different data selection methods are adopted to select meaningful new samples for model update.Moreover,recursive mechanism is introduced to reduce the computation amount.Then,interactive multiple-model method is utilized to combine the two models for prediction.The simulation results show that the proposed method can reduce the errors form prediction with single model.Aiming at the complex distributed parameters characteristic and the uncertainty of geologic parameters,we propose the spatiotemporal decomposition model for ASP flooding based Gaussian process.Firstly,KL decomposition method is used to separate the water saturation,the dominant spatial basis functions and corresponding time coefficients can be obtained.With the injection concentration as input and time coefficients as output,the time identification model can be built by Gaussian process,where the improved KH algorithm is employed to optimize the hyper-parameters.Then,the dynamic model between injection concentration and water saturation can be obtained.Finally,the complete spatiotemporal model for ASP flooding is obtained by identifying the relationship between water saturation and moisture content.We utilize the numerical modeling software CMG to get the reservoir data,the simulation results show the effectiveness of proposed model.Aiming at the injection optimization problems for ASP flooding,the dynamic programming model is established with the net present value maximization as goal.Firstly,the two mentioned models for ASP flooding are compared and analyzed,the spatiotemporal decomposition model is selected to establish the dynamic programming model for ASP flooding.Then,iterative dynamic programming method is employed to solve the model,at the meanwhile,the results are compared with those by improved KH algorithm.The simulation results show the proposed method can improve the net present value efficiently.