Dynamic Programming For Polymer Flooding In Enhanced Oil Recovery With Variable Oil Price

Polymer flooding is an important tertiary oil recovery technology in enhanced oilrecovery,which has been widely used in China’s eastern oil fields. The polymer is anexpensive chemical product, the injection process is long and complex, so in order to improveefficiency, the development of scientific and rational control strategy has important practicalsignificance.This paper solved the Kring polymer flooding optimal control problem based ondynamic programming method.In traditional polymer flooding model the price of crude oil is a fixed value, taking intoaccount the current oil price changes frequently and large in magnitude, in this article we treatthe oil price as a variable,and the oil price model was studied.Three models were built basedon different time series analysis method. In the linear regression model we got the10-orderautoregressive oil price forecast model through the analysis of the prediction error.In thethree-layer BP neural network model, the input layer contains four variables, and the outputlayer of a variable. Established oil price model combinaed with linear parts and volatilitycomponent based on the oil price decomposition, and the model has been improved based onthe window scroll of the model. The prediction result of the same historical data indicate thatthe improved composition model has the best predicting results.Built kriging polymer flooding model based on the statistical point of view. In thismodel,the performance index function is the net present value (NPV),the control variables arethe injection concentration for each injection well, the state variable are the moisture contentof the produced fluid in the production wells.Obtained moisture curve exploded modelthrough the Trigonometric approximation of moisture curve.Thereby moisture curve ischaracterized by a set of Fourier coefficients, and regard it as the observation value of themoisture content.In Kriging model, the sample points are the injection concentration of theinjection well, the observed value are the Fourier coefficients for the moisture curves. Thesample were generated randomly by Latin hypercube sampling method, in this way the sample points have a good space-filling.The Kriging polymer flooding model with variable oil price was solved throuth thedynamic programming method.For the control constraints and state constraints in themodel,we transform these constrains into unconstrained control problem by adding penaltyfunctions to the performance index function. To verify the correctness of the Kriging floodingmodel,a reservoir consisting of a production well and4injection wells is studied,the resultsshow that the optimal strategy can improve the effectiveness of polymer flooding.